Wafer-level heterogeneous integration of MEMS actuators Stefan Braun MICROSYSTEM TECHNOLOGY LABORATORY SCHOOL OF ELECTRICAL ENGINEERING ROYAL INSTITUTE OF TECHNOLOGY ISBN 978-91-7415-493-1 ISSN 1653-5146 TRITA-EE 2010:002 Submitted to the School of Electrical Engineering KTH—Royal Institute of Technology, Stockholm, Sweden, in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy Stockholm 2010 ii Wafer-level heterogeneous integration of MEMS actuators Copyright © 2010 by Stefan Braun All rights reserved to the summary part of this thesis, including all pictures and ﬁgures. No part of this publication may be reproduced or transmitted in any form or by any means, without prior permission in writing from the copyright holder. The copyrights for the appended journal papers belong to the publishing houses of the journals concerned. The copyrights for the appended manuscripts belong to their authors. Printed by Universitetsservice US AB, Stockholm 2010. Thesis for the degree of Doctor of Philosophy at the Royal Institute of Technology, Stockholm, Sweden, 2010. ABSTRACT iii Abstract This thesis presents methods for the wafer-level integration of shape memory alloy (SMA) and electrostatic actuators to functionalize MEMS devices. The integration methods are based on heterogeneous integration, which is the integration of diﬀerent materials and technologies. Background information about the actuators and the integration method is provided. SMA microactuators oﬀer the highest work density of all MEMS actuators, however, they are not yet a standard MEMS material, partially due to the lack of proper wafer-level integration methods. This thesis presents methods for the wafer-level heterogeneous integration of bulk SMA sheets and wires with silicon microstructures. First concepts and experiments are presented for integrating SMA actuators with knife gate microvalves, which are introduced in this thesis. These microvalves feature a gate moving out-of-plane to regulate a gas ﬂow and ﬁrst measurements indicate outstanding pneumatic performance in relation to the consumed silicon footprint area. This part of the work also includes a novel technique for the footprint and thickness independent selective release of Au-Si eutectically bonded microstructures based on localized electrochemical etching. Electrostatic actuators are presented to functionalize MEMS crossbar switches, which are intended for the automated reconﬁguration of copper-wire telecommunication networks and must allow to interconnect a number of input lines to a number of output lines in any combination desired. Following the concepts of heterogeneous integration, the device is divided into two parts which are fabricated separately and then assembled. One part contains an array of double-pole single-throw S-shaped actuator MEMS switches. The other part contains a signal line routing network which is interconnected by the switches after assembly of the two parts. The assembly is based on patterned adhesive wafer bonding and results in wafer-level encapsulation of the switch array. During operation, the switches in these arrays must be individually addressable. Instead of controlling each element with individual control lines, this thesis investigates a row/column addressing scheme to individually pull in or pull out single electrostatic actuators in the array with maximum operational reliability, determined by the statistical parameters of the pull-in and pull-out characteristics of the actuators. Keywords: Microelectromechanical systems, MEMS, silicon, wafer-level, integration, heterogeneous integration, transfer integration, packaging, assembly, wafer bonding, adhesive bonding, eutectic bonding, release etching, electrochemical etching, microvalves, microactuator, Shape Memory Alloy, SMA, NITINOL, TiNi, NiTi, cold-state reset, bias spring, stress layers, crossbar switch, routing, switch, switch array, electrostatic actuator, S-shaped actuator, zipper actuator, addressing, transfer stamping, blue tape Stefan Braun, [email protected] Microsystem Technology Laboratory, School of Electrical Engineering KTH—Royal Institute of Technology, SE-100 44 Stockholm, Sweden iv Wafer-level heterogeneous integration of MEMS actuators ... I have not ﬁlled this volume with pompous rhetoric, with bombast and magniﬁcent words, or with the unnecessary artice with which so many writers gild their work. I wanted nothing extranous to ornament my writing, for it has been my purpose that only the range of material and the gravity of the subject should make it pleasing. ... From “Il Principe” Niccolò Machiavelli, 1469-1532 Translated by Peter Constantine, “The Prince - A new translation” ABSTRACT v To my family vi Wafer-level heterogeneous integration of MEMS actuators CONTENTS vii Contents Abstract iii List of papers ix 1 Introduction and structure 1 2 MEMS actuators 2.1 Common microactuation mechanisms . . . . . . . . . . . . . . . . 2.2 Shape Memory Alloy actuation . . . . . . . . . . . . . . . . . . . 2.2.1 Shape Memory Eﬀects . . . . . . . . . . . . . . . . . . . 2.2.2 Actuation aspects of Titanium-Nickel alloys . . . . . . . . 2.3 Electrostatic actuation . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Parallel-plate, comb-drive and curved-electrode actuators 2.3.2 S-shaped ﬁlm actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 5 5 7 10 11 13 3 Heterogeneous integration 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Monolithic and hybrid integration . . . . . . . . . . . . . . . 3.1.2 Heterogeneous integration . . . . . . . . . . . . . . . . . . . . 3.2 Heterogeneous integration concepts and technologies . . . . . . . . . 3.2.1 Transfer/direct and wafer-to-wafer/chip-to-wafer integration . 3.2.2 Electrical interconnection . . . . . . . . . . . . . . . . . . . . 3.2.3 Wafer-bonding techniques . . . . . . . . . . . . . . . . . . . . 3.2.4 Releasing structures for actuation . . . . . . . . . . . . . . . . . . . . . . . 15 15 15 17 19 19 21 22 24 4 Knife gate microvalves with bulk SMA microactuators 4.1 Integration of Titanium-Nickel shape memory alloy . . . . 4.2 Gas microvalves . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Background . . . . . . . . . . . . . . . . . . . . . . 4.2.2 Knife gate microvalves . . . . . . . . . . . . . . . . 4.3 TiNi sheet actuated knife gate valves . . . . . . . . . . . . 4.3.1 Design . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Fabrication - Integration of TiNi sheets . . . . . . 4.3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . 4.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 27 30 30 34 35 35 35 37 37 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii Wafer-level heterogeneous integration of MEMS actuators 4.4 4.5 TiNi wire actuated knife gate valves . . . . . 4.4.1 Concept . . . . . . . . . . . . . . . . . 4.4.2 Fabrication - Integration of TiNi wires 4.4.3 Results . . . . . . . . . . . . . . . . . 4.4.4 Discussion . . . . . . . . . . . . . . . . Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 38 39 39 39 41 5 Automated main distributing frames with S-shaped actuator switches 43 5.1 Switch units in automated main distributing frames . . . . . . . . . . 43 5.2 MEMS switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 5.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 5.2.2 The S-shaped actuator switch . . . . . . . . . . . . . . . . . . . 47 5.3 The MEMS crossbar switch . . . . . . . . . . . . . . . . . . . . . . . . 48 5.3.1 The routing network . . . . . . . . . . . . . . . . . . . . . . . . 48 5.3.2 The crosspoint switches . . . . . . . . . . . . . . . . . . . . . . 50 5.3.3 The integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 5.3.4 Individual switch addressing . . . . . . . . . . . . . . . . . . . . 50 5.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 5.5 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 6 Summaries of the Appended Papers 57 7 Conclusions 61 Acknowledgements 63 References 65 Paper reprints 79 LIST OF PAPERS ix List of papers The presented thesis is based on the following journal papers: 1. Out of plane knife gate microvalves for controlling large gas ﬂows S. Haasl, S. Braun, A. S. Ridgeway, S. Sadoon, W. van der Wijngaart and G. Stemme IEEE/ASME Journal of Microelectromechanical Systems, vol. 15, no. 5, pp. 1281– 1288, Oct. 2006. 2. Wafer-scale manufacturing of bulk shape memory alloy microactuators based on adhesive bonding of Titanium-Nickel sheets to structured silicon wafers S. Braun, N. Sandström, G. Stemme and W. van der Wijngaart IEEE/ASME Journal of Microelectromechanical Systems, accepted for publication 3. Design and wafer-level fabrication of SMA wire microactuators on silicon D. Clausi, H. Gradin, S. Braun, J. Peirs, G. Stemme, D. Reynaerts and W. van der Wijngaart IEEE/ASME Journal of Microelectromechanical Systems, accepted for publication 4. Localized removal of the Au-Si eutectic bonding layer for the selective release of microstructures H. Gradin, S. Braun, G. Stemme and W. van der Wijngaart IOP Journal of Micromechanics and Microengineering, vol. 19, no. 10, pp. 105014– 105023, Oct. 2009. 5. Single-chip MEMS 5×5 and 20×20 double-pole single-throw switch arrays for automating telecommunication networks S. Braun, J. Oberhammer and G. Stemme IOP Journal of Micromechanics and Microengineering, vol. 18, no. 1, pp. 015014– 015025, Jan. 2008. 6. Row/Column addressing scheme for large electrostatic actuator MEMS switch arrays and optimization of the operational reliability by statistical analysis S. Braun, J. Oberhammer and G. Stemme IEEE/ASME Journal of Microelectromechanical Systems, vol. 17, no. 5, pp. 1104– 1113, Oct. 2008. The contribution of Stefan Braun to the diﬀerent publications: 1 2 3 4 5 6 part of fabrication, all experiments, part of writing major part of design, major parts of fabrication and experiments, all writing part of design, fabrication, experiments and writing part of design, fabrication and experiments, major part of writing major part of design, all fabrication and experiments, major part of writing major part of design, all fabrication and experiments, major part of writing x Wafer-level heterogeneous integration of MEMS actuators The work has also been presented at the following international conferences: 1. Small footprint knife gate microvalves for large ﬂow control S. Braun, S. Haasl, S. Sadoon, A. S. Ridgeway, W. van der Wijngaart and G. Stemme The 13th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems (TRANSDUCERS), Seoul, Korea, June 2005, pp. 329–332. 2. MEMS single chip microswitch array for re-conﬁguration of telecommunication networks S. Braun, J. Oberhammer and G. Stemme Proceedings of the 36th European Microwave Conference (EUMW), Manchester, UK, Sep. 2006, pp. 811–814. 3. MEMS single-chip 5x5 and 20x20 double-switch arrays for telecommunication networks S. Braun, J. Oberhammer and G. Stemme 20th IEEE Int. Conf. on Micro Electro Mechanical Systems (MEMS), Kobe, Japan, Jan. 2007, pp. 811–814. 4. Smart individual switch addressing of 5x5 and 20x20 MEMS double-switch arrays S. Braun, J. Oberhammer and G. Stemme The 14th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems (TRANSDUCERS), Lyon France, June 2007, pp. 153–156. 5. Robust trimorph bulk SMA microactuators for batch manufacturing and integration S. Braun, T. Grund, S. Ingvarsdottír, W. van der Wijngaart, M. Kohl and G. Stemme The 14th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems (TRANSDUCERS), Lyon, France, June 2007, pp.2191–2194. 6. Wafer-scale manufacturing of robust trimorph bulk SMA microactuators N. Sandström, S. Braun, T. Grund, G. Stemme, M. Kohl and W. van der Wijngaart Proceedings of the 11th Int. Conf. on new Actuators (ACTUATOR), Bremen, Germany, June 2008, pp. 382–385. 7. Microactuation utilizing wafer-level integrated SMA wires D. Clausi, H. Gradin, S. Braun, J. Peirs, G. Stemme, D. Reynaerts and W. van der Wijngaart 22nd IEEE Int. Conf. on Micro Electro Mechanical Systems (MEMS), Sorrento, Italy, Jan. 2009, pp. 1067–1070. 8. Selective electrochemical release etching of eutectically bonded microstructures H. Gradin, S. Braun, M. Sterner, G. Stemme and W. van der Wijngaart The 15th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems (TRANSDUCERS), Denver, USA, June 2009, pp. 743–746 LIST OF PAPERS xi 9. Full wafer integration of shape memory microactuators using adhesive bonding N. Sandström, S. Braun, G. Stemme and W. van der Wijngaart The 15th IEEE Int. Conf. on Solid-state Sensors, Actuators and Microsystems (TRANSDUCERS), Denver, USA, June 2009, pp. 845–848 xii Wafer-level heterogeneous integration of MEMS actuators 1 1 INTRODUCTION AND STRUCTURE 1 Introduction and structure This thesis presents research in the ﬁeld of microelectromechanical systems (MEMS), also referred to as micromachines in Japan or Micro Systems Technology (MST) in Europe. MEMS technology uses the tools and techniques that were developed for the Integrated Circuit (IC) industry and allows for high volume parallel production of devices, potentially resulting in low fabrication costs per device. MEMS technology includes components with typical sizes between 1 to 100 μm (1 μm = 0.001 mm), which are combined to form MEMS devices such as pressure sensors, inertial sensors, switches, pumps, valves and many more with dimensions in the mm range. While IC devices can be considered as the ’brain’ of a microsystem, MEMS devices provide the ’arms’ and the ’eyes’ to allow the IC device to sense and manipulate the environment. This thesis focuses on the ’arms’, i.e. actuators which typically convert electrical energy into mechanical movement. Methods were developed to integrate actuators with silicon structures to fabricate microvalves for controlling large gas ﬂows and crossbar switches for automating parts of copper-wire telecommunication networks. The actuators are not integrated using the conventional monolithic fabrication, but using concepts of heterogeneous integration. Thus, the actuators are fabricated separately and ﬁnally bonded onto the silicon structures to functionalize. This thesis is divided into two parts. The ﬁrst part provides detailed background information and informative references to facilitate a better understanding of the second part, which contains the appended journal applications. The ﬁrst part contains four chapters. In chapter 2, common MEMS actuator technologies are introduced with focus on electrostatic and shape memory actuation, which are crucial elements of the MEMS devices presented later on. Chapter 3 introduces heterogeneous integration with reference to the other integration methods, which are monolithic and hybrid integration. The chapter provides background information about heterogeneous integration concepts and technologies including wafer-bonding methods and releasing of structures to be manipulated by integrated actuators. Chapter 4 introduces the concept of knife gate gas microvalves, methods to heterogeneously integrate bulk shape memory sheets and wires for actuation and the combination of both. Other microvalve types and shape memory integration methods are discussed. First prototypes are presented and their performance is compared to other microvalves with respect to performance per consumed silicon footprint area. The work is discussed and an outlook on future work is given. Chapter 5 introduces a MEMS crossbar switch for automating parts of copper-wire telecommunication networks. The chapter presents in detail the application, the integration of arrays of electrostatically actuated switches and a method to individually actuate one out of the 400 switches in the array without underlying CMOS addressing circuits. The work is discussed and an outlook on future work is given. 2 Wafer-level heterogeneous integration of MEMS actuators 2 MEMS ACTUATORS 2 3 MEMS actuators Actuators and sensors are transducers which convert one type of energy into another one. Sensors are mostly used to measure a physical parameter and report it in form of an electrical signal. Actuators work the other way, typically converting electrical energy into mechanical work output. The term ’MEMS actuators’ summarizes all actuators which typically are submillimeter sized and fabricated with MEMS technologies. The following chapters will provide a short summary of the main MEMS actuation principles, with focus on electrostatic and shape memory actuation which are important parts of this thesis. 2.1 Common microactuation mechanisms MEMS technology oﬀers a wide range of microactuators which can be classiﬁed into electrostatic, thermal, piezoelectric, magnetic and shape memory alloy actuation methods. Table 1 and the following paragraphs present a brief overview of the principles according to . A more detailed review of all the diﬀerent microactuators and their principles would go beyond the scope of this thesis and the interested reader is referred to literature for deeper information [1, 2, 3, 4, 5, 6, 7]. The piezoelectric actuation utilizes the coupling of mechanical deformation and electric polarization in certain materials. When applying a mechanical stress to the material, it generates an electrical voltage. This eﬀect is called direct piezoelectric eﬀect and is used for sensing and energy harvesting applications. However, for actuation the inverse piezoeﬀect is deployed: by applying a voltage the material generates mechanical movements. The stroke of piezoelectric actuators is in general very small, however, relatively large forces can be obtained with a very precise displacement resolution. Furthermore, these actuators are fast, allowing for high cycling frequencies and making them very feasible for repetitive actuation. An example of a commercial application deploying the advantages of piezoelectric actuation is the smallest Piezo LEGS® linear motor, which allows for linear travel distances limited only by the length of the displaced element (comes with a 50 mm long element) at a speed of 20 mm s with 10 N force and a resolution smaller than 1 nm. For MEMS applications, the integration of piezoelectric actuators is challenging. After the deposition the ceramic material, which most commonly is lead zirconate titanate (PZT), must be sintered at high temperatures (> 600 °C), which limits the 4 Wafer-level heterogeneous integration of MEMS actuators Table 1. The diﬀerent principles commonly used for microactuation. The real work density might be substantially lower. Method Principle Work density Electrostatic Attractive force between bodies with diﬀerent electrostatic charges. ≈ 105 Piezoelectric Shape change under an electric ﬁeld (inverse piezoeﬀect). ≈ 1.2·105 Thermal Thermal expansion of single material (includes sealed liquid) or diﬀerence in CTE between two materials or phase change. ≈ 5·106 J m3 Magnetic Interaction with magnetic ﬁelds. ≈ 4·105 J m3 Shape memory alloy Temperature dependent crystal phase transformation with macroscopic shape change. Belongs technically to thermal actuation. ≈ 107 J m3 J m3 J m3 range of compatible materials and processes. Furthermore, the ceramic shrinks during sintering, resulting in very high mechanical stresses if the ceramic is constrained by the substrate prior to sintering. Magnetic actuation is based on interaction with magnetic ﬁelds which are generated by either permanent magnets or coils. The energy density of magnetic actuation is in the same order of magnitude as electrostatic actuation. For some applications magnetic actuation is considered as an alternative to electrostatic actuation because of advantages such as non-existence of electric discharges, possibility of operation in liquid and long-range forces. However, integrating magnetic ﬁeld sources or low reluctance materials on-chip is a major challenge; the coils are three-dimensional structures which are complicated to fabricate and the hard or soft magnetic materials are diﬃcult to integrate with MEMS structures . Thermal actuation is based on volume or phase change of materials upon heating or cooling. The displacement is analog to temperature change and thereby sensitive to environmental inﬂuences. Typically, thermal actuators provide large stroke and large forces, however, with a rather limited displacement resolution. Heating of the actuators can be very fast by a high applied power, however, cooling is in most cases passive and the cooling time severely slows down the actuator and decreases the actuation frequency. Examples of MEMS thermal actuation approaches are the ’heatuator’  using one material, bimorph using diﬀerent materials , thermopneumatic actuation and shape memory alloys. The ’heatuator’ is a U-shaped lateral thermal actuator fabricated in one material, with one arm of the ’U’ considerably narrower than the other. Current is passed through the actuator, and the higher current density in the narrower 2 MEMS ACTUATORS 5 ’hot’ arm causes it to heat and expand more than the wider ’cold’ arm. The arms are joined at the free end, which constrains the actuator tip to move laterally in an arcing motion towards the cold arm side. The bimorph actuation scheme is based on the diﬀerence in coeﬃcients of thermal expansion (CTE) between two joined layers of diﬀerent materials. When changing the ambient temperature, one layer expands or shrinks more than the other and the resulting interface stress causes bending of the stack. The amount of bending can be controlled by choosing the appropriate CTE combination and by the applied temperature/electrical power. The thermopneumatic actuation scheme is based on the thermal expansion of a ﬂuid sealed inside a cavity. Heating causes volume expansion or even phase change, exerting a large force on the cavity walls and causing a bending of a deformable membrane. Shape memory alloys are materials which undergo a phase change upon temperature changes. This phase change comes with a macroscopic shape change of the device and provides the highest energy density of all MEMS actuators. SMA actuation is an important part of this thesis and is presented in detail in the next subsection. 2.2 Shape Memory Alloy actuation Technically, shape memory alloy (SMA) actuators are thermal actuators, since thermal energy triggers a crystal phase change in the material. However, since SMA actuation is an important part of this thesis, it is presented on its own. 2.2.1 Shape Memory Eﬀects Structures made from Shape Memory Alloys exhibit the shape memory eﬀect (SME), which is the ability of certain materials to ’remember’ their initial shape after they have been deformed. Take a spring made out of SMA and pull it. It will be easy to deform and it will stay deformed. Now heat the spring above a speciﬁc transformation temperature and it will rapidly recover the initial shape. The underlying mechanism of the shape memory phenomenon is a martensite to austenite and vice versa phase transformation. Figure 1 illustrates the diﬀerent crystal states and their connected macroscopic shapes by means of a SMA spring. In the hot state, the SMA crystal structure thermodynamically prefers a more ordered phase and transforms to the austenite crystal form with the macroscopic shape connected to it. This phase is also called parent phase and it is possible to set the desired macroscopic shape by a speciﬁc treatment involving mechanical constraining and heating. When the spring is cooled again and there is no external force applied, it will still have the same shape as in the hot state, however, its crystal is not in the cubic form anymore. Instead, the layers in the material are tilted, with the tilting direction alternating between each layer. Because of these alternated tilts, the spring remains its shape even though the crystal form has changed. The material of the spring is now in a macroscopically non-deformed, low-temperature phase referred to as selfaccommodated martensite  or ’twinned’ martensite , since its characteristic alternated layers are called ’twins’. In this phase, the material features a very low yield strength and can easily be plastically deformed after straining it above a very narrow elastic strain range - therefore the spring can be pulled very easily. After the external force is removed, the spring will stay deformed except for a very small elastic strain 6 Wafer-level heterogeneous integration of MEMS actuators austenite phase parent phase at in g he g in ME at he ay S w otw co o lin g macroscopic shape ad lo y a d it lo tic A f un elas > T rr pe fo g su nly in o ol co crystal phase g E in at SM he a y w eon crystal phase macroscopic shape self-accomodated martensite twinned martensite crystal phase mechanical load macroscopic shape detwinned martensite Figure 1. Illustration of the transformation processes in martensitic transformation, including all the shape memory eﬀects. recovery. During the deformation, the alternated layers of the twinned martensite are moved and the macroscopic shape of the structure is changed. Now the material is still in a martensitic phase and easily deformable. Since the alternated layers called twins are removed, this martensitic phase is called ’detwinned’ martensite. Figure 2 displays the hysteresis behavior of the phase transformation versus the temperature. The transformation from martensite (cold) to austenite (hot) starts at the austenite start temperature As and ﬁnishes at the austenite ﬁnish temperature Af . Vice versa, the transformation from austenite (hot) to martensite (cold) starts at the martensite start temperature Ms and ﬁnishes at the martensite ﬁnish temperature Mf . Hence, there is no single transformation temperature. However, in practice it is referred to a temperature T0 , which is the thermodynamic equilibrium temperature of martensite and austenite state (T0 = MS 2+As ) . There are three diﬀerent shape memory eﬀects, which are illustrated in the overview in ﬁgure 1. In the example above, the spring must be deformed with an external force and performs work only in one direction from detwinned martensite to austenite. Thus, this eﬀect is called the one-way eﬀect. However, the material can be trained to assume a certain shape in the cold state [12,13]. Then, the crystal transforms directly between austenite and detwinned martensite. Hence, the material performs work in two directions and this eﬀect is called the two-way eﬀect. The third eﬀect is called superelasticity or pseudoelasticity and is only present if the SMA is always at temperatures above the austenite ﬁnish temperature Af . Then, if an externally applied stress overcomes a critical stress, the crystal transforms from austenite to detwinned 2 MEMS ACTUATORS 7 crystal phase austenite martensite temperature T Mf Ms T0 As Af Figure 2. Illustration of the hysteresis behavior of the martensitic transformation with the associated temperatures: austenite start As , austenite ﬁnish Af , martensite start Ms and martensite ﬁnish Mf . The temperature T0 is the thermodynamic equilibrium temperature of martensite and austenite state. martensite. When the stress is removed, the crystal immediately transforms back to austenite. In this case, the spring in the example above will immediately, and without extra energy supply, recover the straight shape when the deforming external force is removed. 2.2.2 Actuation aspects of Titanium-Nickel alloys The shape memory phenomenon was ﬁrst discovered in the 1930s in brass alloys. In 1962, Buehler and his colleagues found the shape memory eﬀect in alloys of Nickel and Titanium  and in honor of their employer they named these alloys NiTiNOL (Nickel Titanium Naval Ordinance Laboratory). Nowadays, these allows are also known under the acronyms TiNi (Titanium-Nickel) or NiTi (Nickel-Titanium). In this thesis, the term TiNi is used. Besides TiNi, there are a number of other materials showing the shape memory eﬀect, such as other metallic alloys, polymers and even bacteria [15, 16]. However, TiNi based SMA devices are dominating the market because of several advantages over other alloy systems. TiNi alloys allow to adjust the transformation temperature T0 over a wide range only by changing the ratio of nickel atoms. If the alloy consists of half nickel and half titanium atoms, the transformation occurs near 100 °C. However, adding slightly more nickel atoms decreases the transformation temperature to below 0°C. Furthermore, these alloys can be fabricated with standard metalworking techniques, they exhibit better shape memory strain performance than other known alloys and consist of the aﬀordable elements Nickel and Titanium. TiNi is a biocompatible material, making it interesting especially for medical applications. All the following reﬂections and work presented in this thesis are based on TiNi alloy. SMA actuation is generally based on the one-way eﬀect, i.e. when heated upon deformation the structure recovers its initial shape, yet upon cooling the shape does not change by itself. The approaches to utilize the one-way SME are usually summarized in three diﬀerent categories, depending on the load which is applied upon the SMA during shape recovery [5, 11, 17]. 1. In free recovery, a deformed SMA device is not constrained by any external load during the shape recovery and therefore the SMA does not provide any force. 8 Wafer-level heterogeneous integration of MEMS actuators 2. In constrained recovery, the shape recovery of a deformed SMA device is blocked by an external constraint, triggering large forces from the SMA. 3. In cyclic work production, the shape recovery is constrained. However, upon heating the SMA can overcome the external force for the shape recovery. Upon cooling, the external force deforms the SMA again until the next temperature cycle. The external force is called bias spring or cold-state reset. The combination of SMA and bias spring described under cyclic work production is the basis for SMA microactuators. The methods providing the cold-state reset can be summarized as intrinsic and extrinsic methods . Using intrinsic methods, the crystal of the material is modiﬁed to prefer a certain cold-state crystal orientation, which results in a preferred shape of the structure in the cold-state. An example of an intrinsic cold-state reset is the two-way shape memory eﬀect [19, 20], where a cold-state shape is trained into the material using long-term cycling processes. However, compared to extrinsic cold-state reset methods, the twoway eﬀect is very unstable [19, 20], exhibits considerably less recoverable deformation and furthermore the required training process is diﬃcult to integrate with a batch fabrication process for MEMS applications [5, 21]. Therefore this method will not be further discussed in this thesis. Most of the SMA actuators deploy extrinsic biasing methods, where the SMA material is coupled with an additional mechanical element. A widely used biasing scheme, especially in MEMS applications, is coupling of the SMA with an external biasing spring element. This topic will be addressed in detail below. Another interesting scheme is the antagonistic biasing, where two SMA elements are coupled together. While element A remains cold and very easy to deform, element B is heated and pulls the cold element A without requiring high forces. Then, after cooling, both elements maintain their current shape until element A is heated, thereby deforming the cold and soft element B. Since the SMA bias spring is very easy to deform over relatively large strains (of course only within the elastic range), large deﬂections can be obtained. Yet, in MEMS applications it is diﬃcult to couple two SMA elements in combination with a good thermal isolation between them. The achievable work density of the TiNi is very much depending on the load case. Table 2  shows the diﬀerent cases. Under pure tension or compression load, the highest forces can be obtained, however, with relatively small displacements. Larger displacements, yet with lower forces, can be obtained under torsion or bending load. Bending load provides the lowest energy density of the three diﬀerent load cases, since only fractions of the material are used for work production. Another important aspect to consider when designing the actuator is the fatigue of the material. Fatigue in TiNi is usually divided into structural and functional fatigue . Structural fatigue refers to the mechanical failure of the TiNi after cyclic loads, similar to any other engineering material. But unlike normal engineering materials, shape memory alloys show diﬀerent properties in diﬀerent temperature ranges, which also inﬂuences the fatigue characteristics. Functional fatigue refers to a decrease in functional properties, which is the shape 2 MEMS ACTUATORS 9 Table 2. Comparison of work density and energy eﬃciency of TiNi wires for three diﬀerent load cases . Load case J Work density ( kg ) Energy eﬃciency (%) Tension/Compression Torsion Bending 466 82 4.6 1.3 0.23 0.013 Table 3. Allowable stress and strain for a targeted amount of actuation cycles . Cycles Max.strain (%) Max. stress (MPa) 1 102 104 > 105 8 4 2 1 500 275 140 70 recovery of the TiNi during cyclic loading. The functional fatigue is of great interest, since it deﬁnes how many cycles the actuator can be operated depending on the stress and strain applied to the material. Table 3  shows some experimentally evaluated benchmark numbers for TiNi wires. When straining the wire with the maximum possible 8% or stressing it with the maximum possible 500 MPa, only one shape recovery cycle can be obtained. To maximize the number of actuation cycles, the applied strain should be below 1% or the applied stress should be below 70 MPa. Thermal energy must be provided to trigger the shape recovery of the SMA. An option is to vary the ambient temperature. However, for cyclic actuation purposes this is rather impractical. The TiNi can be heated by electrically contacting and Joule heating the material itself. However, the stable oxide on the TiNi makes electrical contacting complicated. Therefore, especially for MEMS applications, the heating is sometimes performed indirectly using a separate resistive heater, which can be contacted in a simpler way. The voltages needed to operate SMA microactuators are compatible with microelectronics, however, high currents are necessary to provide the relatively high power for heating of the SMA. During operation, the TiNi transforms between two states and displays hysteresis behavior as described earlier. Because of the two stable states, the SMA is very suitable for applications that require digital mode operation of the actuator. For such applications, the hysteresis behavior is potentially of advantage; the thermal energy necessary to maintain the austenite state is lower than the initial austenite start temperature As , which deﬁnes the stable state very well even for an unstable thermal energy supply. For applications requiring precise analog-like control over the displacement of the actuator, there are several controlling solutions. One method is the model-based loop, which is based on extensive modeling of the materials behavior to reduce or compensate the hysteresis eﬀect. However, the necessary material param- 10 Wafer-level heterogeneous integration of MEMS actuators pulled maintains current out state d k Fs moveable plate, Area A gradual deflection d0 pull-in 2d 3 0 d0 d Fel V pulled in pull-out fixed plate Vpull-out (b) (a) Vpull-in V (c) Figure 3. Diagram illustrating the operational behavior of an electrostatic actuator: (a) and (b) Parallel-plate capacitor model showing the principle of electrostatic actuation with and without applied voltage, respectively; (c) Diagram illustrating the hysteresis behavior of an electrostatic actuator, showing the typical pull-in and pull-out characteristic. eters must be experimentally evaluated. Another method, the feedback-loop, relies on sensing of either position , temperature  or electrical resistance  of the SMA structure as input. Both position and temperature sensing require additional devices and especially the temperature sensing is impractical due to temperature disturbances in an open environment. The control based on electrical resistance sensing is very interesting since it utilizes the smart material capabilities of the SMA. Yet another interesting approach is to keep the digital mode operation with all its advantages, but segmenting the SMA element to quantify the deﬂection . This overview is far from being complete, therefore the interested reader is referred to  for more information on this topic. 2.3 Electrostatic actuation The electrostatic actuation principle relies on the attraction force between bodies having diﬀerent electrostatic potential caused by a charge inbalance. A simple example of an electrostatic actuator is a parallel-plate capacitor, as illustrated in Figure 3a, with one ﬁxed plate and the other plate suspended by a mechanical spring with a spring constant k at an initial distance d0 . Applying a voltage V between the two plates results in a vertically attractive electrostatic force, which pulls the moveable plate towards the ﬁxed plate (Figure 3b). Using this simpliﬁed model and neglecting fringe-ﬁelds, the electrostatic force Fel between the plates can be calculated as  Fel = A 1 ε0 r 2 V 2 2 d (1) with ε0 the permittivity of free space, r the eﬀective relative permeability, A the overlap area of the two plates and d the distance between the two plates. This formula is the basic formula for all electrostatic actuators and shows that the electrostatic force grows quadratically with decreasing distance between the plates, which makes electrostatic actuation very interesting for MEMS applications with very small gap distances below tens of micrometers. 2 MEMS ACTUATORS 2.3.1 11 Parallel-plate, comb-drive and curved-electrode actuators For electrostatic actuators based on the parallel-plate concept there are several issues to consider. The electrostatic force Fel is counteracted by the mechanical spring force Fs , which is calculated as: Fs = −k(d0 − d) (2) For practical designs with a low actuation voltage, small electrode area and a suﬃciently stiﬀ mechanical spring, the initial plate distance d0 must be small, which results in small travel distances d (strokes) of typically a few micrometers for the moveable plate. Furthermore, the range in which the stroke of the moveable plate can be controlled is limited. Figure 3c illustrates the deﬂection of the moveable plate during a full operation cycle. With increasing actuation voltage, the gap between the plates gradually decreases and the two forces Fel and Fs will settle in an equilibrium. However, with decreasing d, the electrostatic force grows quadratically, whereas the counteracting spring force only grows linearly. At distances smaller than the critical distance d = 23 d0 , there no longer exists an equilibrium between the forces and the moveable plate snaps down to the ﬁxed plate. To avoid an electrical short-circuit after snap-down, there must be an electrical isolation layer or at least ’dimples’ (distance holders) between the plates. The critical distance is independent of the geometrical parameters of the actuator and the voltage at which the plate snaps down is called the pull-in voltage, or Vpull−in (ﬁgure 3c). After the pull-in, the gap d is drastically minimized and therefore, when reducing the applied voltage after pull-in, the electrostatic force remains larger until a force equilibrium is reached again. When further reducing the applied voltage, Fs overcomes Fel and the moveable plate is pulled out. Accordingly, the voltage at which the pull-out occurs is called Vpull−out (ﬁgure 3c). Some applications require an analog behavior of the actuator and there are eﬀorts to extend the limited analog controllable stroke of parallel-plate actuators [30,31,32]. However, there are also many applications demanding a digital mode operation of the actuator, such as electrical and optical switches which alternate between the ON and the OFF state. In these applications, the hysteresis behavior is actually of advantage; the voltage necessary to maintain the pull-in state is lower than the initial actuation/pull-in voltage, which deﬁnes the switching state very well even at unstable control voltages. In contrast to the parallel-plate actuator, the ideal comb-drive actuator shows no pull-in and hysteresis behavior since the plates are not moving perpendicularly, but parallel to each other and thereby keep the plate distance d constant during the operation. A second ﬁxed plate is added and the moveable plate is interdigitated between the two ﬁxed plates with an initial lateral overlap x0 . Figure 4 illustrates the actuator, which is called ’comb-drive’ since the interdigitated ﬁnger-like structures look like the teeth on a comb. Applying an actuation voltage results in several forces, as illustrated in ﬁgure 4b. The two vertical force components Fel,y keep the moveable ﬁnger centered between the stationary ﬁngers and the lateral force component Fel,x pulls the moveable ﬁnger towards the ﬁxed ﬁngers and counteracts the lateral bias spring with the spring 12 Wafer-level heterogeneous integration of MEMS actuators fixed plate, thickness t x0 Fel,y x d moveable plate, thickness t V k Fel,x Fs d Fel,y fixed plate, thickness t (a) (b) Figure 4. Diagram illustrating the operational behavior of an electrostatic comb-drive actuator, without (a) and with (b) applied actuation voltage. constant k. The lateral force Fel,x is independent of the plate overlap and remains constant with increasing plate overlap. The distance between the capacitor plates also remains constant, which allows for large analog controllable stroke, only limited by the elastic range of the bias spring. However, the large stroke comes at a cost. The ﬁngers are usually fabricated by vertical etching into the silicon substrate using deep reactive ion etching (DRIE). As for all electrostatic actuators, the distance d between the ﬁngers should be as small as possible and the electrode area A as big as possible. Consequently, high aspect ratio processes are necessary to produce structures with a minimal distance in between. Since the aspect ratio and the resulting initial distance is limited, the only way to increase the electrode area and the force is the massive parallelization of comb structures, at the cost of silicon footprint. Both large stroke and large force are provided by curved-electrode actuators, which utilize a ﬂexible beam opposite a ﬁxed electrode. Figure 5 illustrates the principle. One end of the ﬂexible beam is clamped with a very short distance to the ﬁxed electrode. One of the two electrodes is a curved electrode, shaped in a way that the electrode distance is gradually increasing from the clamped end to the free end. Upon applying an actuation voltage, the narrow gap at the clamped end results in large forces and the ﬂexible beam is pulled in. As for the parallel-plate actuator, electrical isolation between the beams is necessary to avoid an electrical shortcircuit. The point of pull-in is moving along the ﬁxed electrode in a zipper-like way and therefore these actuators are also referred to as ’zipper-actuators’. Another name is ’touch-mode’ actuator, since these actuators utilize the pull-in and the plates are touching each other only separated by a thin electrical isolation layer or stoppers. The combination of small plate distance at the clamped end, the touching mode with very thin gaps between the electrodes and the large distance at the free end results in large forces and a large stroke, making this actuation scheme very interesting for MEMS applications. For the most common zipper actuators, the moveable part is moving either laterally or vertically, as illustrated in ﬁgure 5 . Besides the orientation of the actuation, the two conﬁgurations also diﬀer in the arrangement of ﬁxed and moveable plate as well as in their fabrication. In the lateral zipper approach, the ﬁxed electrode 2 MEMS ACTUATORS 13 d0 moveable, straight electrode fixed, curved electrode moveable, curved electrode d0 fixed, straight electrode silicon (a) (b) Figure 5. The two most common fashions of zipper actuators: (a) lateral zipper and (b) vertical zipper. The ﬁgure is modiﬁed from . is curved and the moveable electrode is a straight cantilever. The fabrication is fairly simple with one photolithographical mask, vertical etching into the device layer of a SOI wafer and sacriﬁcially underetching the buried oxide to release the moveable beam. However, as for the comb-drive actuator, the initial gap and the electrode area are limited by the aspect ratio of the fabrication process. In the vertical zipper approach (illustrated in ﬁgure 5b and 6a), the ﬁxed electrode is straight and the moveable electrode is curved, typically fabricated using surface micromachining. The curvature of the moveable electrode results of a controlled, fabrication process related stress gradient and because of the curvature, these kind of actuators are also called ’curled actuators’. In contrast to the lateral approach, the initial gap at the clamped end can be very narrow by utilizing a thin sacriﬁcial layer and the electrode area can be very large, resulting in large forces at relatively low actuation voltages. However, the stress gradient in the moveable electrode and the resulting spring tension counteracting the electrostatic force is diﬃcult to control. Furthermore, stiction could occur between the large area electrodes in close contact. 2.3.2 S-shaped ﬁlm actuators The stress gradient in the bending electrodes of standard vertical zipper actuators is diﬃcult to control. A large stress gradient results in a spring with a high pretensioning, counteracting the electrostatic force and resulting in larger actuation voltages to pull in the electrode. A thin and soft membrane would decrease the necessary actuation voltages, however, a weak spring cannot provide a reliable pull-out and suspension of the electrode. A solution allowing for a thinner and softer membrane is the incorporation of a second ﬁxed electrode at the free end of the membrane, providing a second zipper actuator, as illustrated in ﬁgure 6b. The resulting double-zipper actuator provides active actuation of the ﬁlm in both directions, allowing for a very ﬂexible membrane with a low stress gradient and thereby potentially reducing the actuation voltage yet still allowing for a large stroke. A MEMS concept of such an actuator is illustrated in ﬁgure 6b. First, a single zipper is fabricated with a thin and ﬂexible membrane, 14 Wafer-level heterogeneous integration of MEMS actuators pre-stressed flexible electrode d0 touch-mode, pull-in ’zips’ along fixed electrode V V fixed electrode (a) single ’zipper’ actuator 2nd fixed electrode two ’zippers’, d0 defined by thin electrical isolation layer between the electrodes d0 fixed electrode V d0 (b) double ’zipper’ or S-shaped film actuator Figure 6. Illustration of (a) single vertical zipper actuator and (b) double vertical zipper or S-shaped ﬁlm actuator. yet still with suﬃcient bending of the free end. Then, the second electrode is added from the top and pushes the membrane in contact with both electrodes and creating the characteristic S-shape of the membrane which inspired the name of the S-shaped actuator. To allow the membrane to move up and down, the two ﬁxed electrodes must be kept at a distance to each other with an intermediate spacer. The thickness of this spacer allows to tune the distance the membrane can move up and down between the two electrodes, which deﬁnes the stroke of the actuator. In summary, this concept comes with a set of advantages. The touch-mode actuation, with very small initial gaps in both directions, in combination with a thin and ﬂexible membrane potentially results in very low actuation voltages. The stroke of the actuator is basically only limited by the tuneable spacing between the two ﬁxed electrodes. The S-shaped actuator was shown in 1997 for a gas valve with dimensions in the millimeter range . Another work [35, 36, 37] utilized the assembly concept of the S-shaped actuators to fabricate an RF MEMS switch by fabricating a single zipper actuator with metal contacts on one substrate and combining it with a second substrate, which contained the second ﬁxed electrode and signal lines to be interconnected. In this work, the spacing between the substrates was provided by a polymer ring, which also encapsulated and packaged the switch. 3 HETEROGENEOUS INTEGRATION 3 15 Heterogeneous integration Heterogeneous integration evolved from monolithic and hybrid integration and refers to the wafer-level integration of diﬀerent materials, technologies or devices. In monolithic integration a device is fabricated in one piece while in hybrid integration a device is fabricated by interconnecting several separate pieces. The following sections introduce the diﬀerent integration methods, followed by technical background including methods for wafer-to-wafer bonding, vertical electrical interconnection and releasing of structures for actuation. 3.1 Introduction The following sections introduce the concepts of monolithic, hybrid and heterogeneous integration. Heterogeneous integration is of high interest for the integration of MEMS and IC and therefore the integration technologies are introduced by means of the speciﬁc example of integrating MEMS materials onto IC circuits. 3.1.1 Monolithic and hybrid integration In monolithic integration, devices are fabricated from one substrate (monolithic = made from one piece). All the processing is typically performed on wafer-level and after the fabrication the wafer is diced into discrete devices (ﬁgure 7a), which are ready for further application. As an example, MEMS and IC are monolithically integrated by combining and customizing the MEMS and IC manufacturing processes. The main technical advantage of monolithic integration of MEMS and IC is the high integration density; electrical interconnections between MEMS and IC are very short, reducing electrical noise and allowing for the handling of small signals. However, monolithic integration of MEMS and IC is relatively complicated [38, 39, 40, 41, 42], since MEMS technology can require IC incompatible material deposition processes and/or temperatures above 450 °C, which is not allowed for the IC components. A solution to avoid these problems is the hybrid integration (hybrid = combination of diﬀerent parts), where the devices are fabricated on separate substrates, which are then diced into single chips and combined with each other on chip level (ﬁgure 7b). As an example, MEMS and IC are hybrid integrated by fabricating on separate substrates, which are then diced into single MEMS and IC chips. Conventionally, 16 Wafer-level heterogeneous integration of MEMS actuators MEMS+IC substrate dicing Device 1 Device 2 Device 3 (a) monolithic integration MEMS substrate IC dicing MEMS 1 substrate dicing MEMS 2 MEMS 3 IC 1 IC 2 IC 3 pick-and-place assembly wirebond adhesive carrier substrate Device 1 carrier substrate Device 2 carrier substrate Device 3 (b) hybrid integration MEMS IC target substrate source substrate bonding removing source substrate dicing Device 1 Device 2 Device 3 (c) heterogeneous integration Figure 7. Simpliﬁed schematic illustrations of the diﬀerent methods for integrating MEMS with IC: (a) monolithic integration, (b) hybrid integration and (c) heterogeneous integration. 3 HETEROGENEOUS INTEGRATION MEMS 17 ASIC wirebonds (a) (b) (c) (d) Figure 8. Example of wire bonding based hybrid integration of a MEMS inertial sensor with an ASIC for automotive applications . (a Leadframe. (b) The MEMS sensor (top) and the ASIC (bottom) are adhesively mounted onto the leadframe. Using wirebonding, the MEMS is electrically connected to the ASIC and the ASIC is connected to the leadframe. (c) Packaged by plastic molding. (d) The leadframe pins are punched free and shaped, resulting in a chip ready for integration in a larger system. these chips are glued beside each other on a carrier substrate and electrically interconnected by wire-bonding (see example in ﬁgure 8) or by connections integrated in the substrate (Multi Chip Modules ). Alternatively, they are stacked on top of each other [43, 44] using through-substrate-vias (TSV) [45, 46, 47, 48] for vertical electrical interconnection. The main technical advantage is the uncomplicated integration of diﬀerent technologies, materials or devices. However, there are applications where hybrid integration is not feasible due to cost-eﬃciency reasons, limited integration density and parasitic signal noise from the long electrical interconnections. Heterogeneous integration allows to combine the two approaches and is presented in the next section. 3.1.2 Heterogeneous integration Typically, heterogeneous integration is utilized for integrating materials or technologies which otherwise are very diﬃcult to combine or even incompatible with each other. However, heterogeneous integration also allows to divide the fabrication of MEMS devices into several separate sub-structures, which are optimized for a certain aspect and ﬁnally combined to one device. Both aspects are included in this thesis. Chapter 4 addresses the integration of an incompatible actuator material with silicon structures and chapter 5 addresses the separate fabrication of MEMS actuators arrays and their integration to functionalize another MEMS device. Heterogeneous integration follows the same basic concept as hybrid integration. The devices to be integrated, or parts of them, are fabricated separately. However, in contrast to hybrid integration, the two diﬀerent substrates are integrated on waferlevel by bonding them on top of each other, followed by removing the substrate of the integrated device and dicing into single chips (ﬁgure 7c). This wafer-level hybrid integration method allows to combine the advantages of hybrid and monolithic integration such as separate fabrication and wafer-level processing including high integration density and short electrical interconnections between the devices. An example to demonstrate all the beneﬁts of heterogeneous integration is the replacement of the mirror material of micromirror arrays from aluminum to monocrys- 18 Wafer-level heterogeneous integration of MEMS actuators fabricate actuation electrodes IC substrate ’dummy IC’ substrate spin-on and pattern sacrificial layer fabricate actuation electrodes spin-on and pattern sacrificial layer SOI - wafer deposit mirror material by sputtering aluminum deposit mirror material by bonding SOI – wafer and removing handle wafer and buried oxide pattern mirror material to form micromirrors pattern mirror material to form micromirrors electroplate vias for electrical interconnection and mechanical clamping remove sacrifial layer (a) monolithic remove sacrifial layer (b) heterogeneous Figure 9. Simpliﬁed illustrations of the fabrication of micromirrors: (a) monolithically fabricated with aluminum mirrors  and (b) heterogeneous integration of a silicon layer for the mirrors . talline silicon. The famous digital micromirror device (DMD) for projectors from Texas Instruments  features an array of up to 2048×1152 micromirrors and each of these mirrors must individually addressable, which for the DMD is performed using a dedicated IC circuit. Using hybrid integration of the mirrors and the IC would not be feasible since more than one million interconnection wires would be necessary to control each mirror. Therefore, the mirrors are monolithically integrated on top of the IC by sputtering and patterning aluminum as mirror material. Figure 9a  shows a simpliﬁed example of such a process. Each mirrors control electrode is addressed by a memory cell directly underneath, which eliminates the need to route individual control wires underneath the array. However, after repeated or prolonged mirror actuation, the aluminum mirrors display hysteresis and memory eﬀects, which can be problematic for applications requiring analog mirror deﬂections. These issues can be eliminated by utilizing monocrystalline silicon for the mirrors. Furthermore, the achievable optical quality, the surface roughness and the uniformity of monocrystalline silicon surfaces is superior compared to most other surfaces. However, since the process temperatures for the material deposition onto IC is limited to about 450 °C to avoid damage to the electronic circuits, many high-performance MEMS materials such as monocrystalline semiconductors cannot be monolithically integrated. A solution is the heterogeneous integration of monocrystalline silicon mirrors onto the IC driving circuitry by transferring the thin silicon device layer of a SOI-wafer using a IC compatible transfer bonding process. The involved temperatures are always below 450 °C. Figure 9b shows a simpliﬁed example of such a process [53,54,55,56,51] 3 HETEROGENEOUS INTEGRATION 19 and the analogy to the monolithic fabrication of micromirror arrays. Heterogeneous integration was enabled with the advent of wafer bonding technologies and is an important technology for the ’More than Moore’ trend to integrate non-IC functions onto IC devices in order to increase their capabilities beyond Moores Law . More information about the technology behind heterogeneous integration is presented in the following section. 3.2 3.2.1 Heterogeneous integration concepts and technologies Transfer/direct and wafer-to-wafer/chip-to-wafer integration The illustration of heterogeneous integration in ﬁgure 7 shows the transfer integration approach, which allows to transfer layers of one substrate to another substrate. These layers can be closed layers or devices which are fabricated in the same layer. The layer to be integrated is ﬁrst fabricated on a temporary carrier substrate, which is the source substrate. The source substrate is then bonded upside down onto the target substrate, followed by removal of the source substrate. As a result, the source layer to be integrated is bonded upside down onto the target substrate. If the element should not be integrated upside down, it can be bonded to another intermediate substrate before it is transferred to the ﬁnal target substrate. An example for transfer integration is the integration of the thin device layer of a SOI wafer onto CMOS substrate for micromirror arrays. In contrast to transfer integration, in direct integration the substrates are bonded directly on top of each other without intermediate carriers or removing substrates. The substrates can be bonded with the top or the bottom side facing each other, depending on the application. Direct integration allows for wafer-scale encapsulation/packaging of devices as illustrated in ﬁgure 10. The source substrate is bonded with the top side onto the top side of the target substrate. The bonding layer is thick and patterned, forming a ring around the devices and encapsulating/packaging the devices after the bonding. Encapsulating MEMS structures by bonding plain wafers with recesses on top of the target wafer is a common wafer-level packaging approach . However, in contrast to only one substrate containing devices, heterogeneous integration allows for both of the substrates containing functional structures. The integration approaches introduced above allow for integrating wafers to wafers, which is typically cost-eﬃcient if the devices on both substrates feature similar footprint areas or if the total cost per footprint of the substrate with the smaller devices is much lower as compared to the ﬁnal total cost of the ﬁnal devices. Figure 11 illustrates this issue: if the device to be integrated features a much smaller footprint area than the device on the target substrate, a lot of expensive substrate material is be wasted. The conventional alternative to integrate devices with largely diﬀerent areas would be the hybrid integration. However, this requires robotic pick-and-place of the components, which is a serial process and especially for high volume production may not be cost-eﬀective. Alternatives based on heterogeneous integration are chip to wafer integration methods, where the source substrate with the source devices is diced into single chips, which are which are bonded onto the target devices on the un-diced 20 Wafer-level heterogeneous integration of MEMS actuators target devices bonding layer devices to be integrated encapsulated/packaged devices rings cross-section top view Figure 10. Schematic illustration of direct integration which results in wafer-level encapsulation/packaging of the integrated devices, based on a thick bonding layer which is patterned to form rings around the devices. target wafer. These methods address the cost issue associated with largely diﬀerent footprint areas and three examples of such techniques are described below. One method is mixing pick-and-place and wafer-level transfer integration. The source substrate is diced into single chips, which are pick-and-place bonded onto the target devices on the target substrate. This approach has been developed by IMECGhent University for the integration of optical chips onto IC substrates , using an intermediate substrate to bond the chips with the correct side to the target substrate. Another approach utilizes self-alignment methods, which are part of self-assembly methods [60, 61, 62, 63]. The smaller devices are densely fabricated on a source substrate which is diced into single chips. These single chips are placed on an assembly wafer, without any orientation or order and due to previous manipulations of the assembly wafer the chips orient themselves along deﬁned patterns with deﬁned pitches. Induced vibration or evaporating liquid helps to overcome the friction between the chips and the surface of the assembly wafer. After the chips have oriented and assembled themselves on the assembly wafer, they can be transferred onto the target wafer. Alternatively, the chips are self-assembled directly on the target wafer. Similar methods have been used to transfer monocrystalline silicon onto surface micromachined electrostatic actuators to fabricate micromirrors . Furthermore, there is a report using self-assembly for the integration of MEMS (actuators) with IC . A third method is the ’selective transfer technology for microdevice distribution’ presented by IBM . Their concept involves the dense fabrication of the smaller devices on a source wafer. This source wafer is aligned to the target wafer in a way, that one small source device is aligned to the larger target device on the target wafer. Then, the source device is released from the source wafer and transferred to the target device. By adapting the pitch of the smaller elements and the target devices, several smaller elements can be transferred in one transfer step. Using this technology, one source wafer can populate a number of target wafers and thereby the 3 HETEROGENEOUS INTEGRATION 21 ”wasted” substrate area source devices target devices (a) (b) Figure 11. Cost-eﬀectiveness of wafer-to-wafer integration: (a) The integration is costeﬀective because the footprint of the transferred device is similar to that of the target device. (b) The wafer-integration is no longer cost-eﬀective due to the large footprint diﬀerences. cost of the transferred wafer is distributed over the number of target wafers. IBM showed this method for the distribution of their AFM-cantilevers and, together with FZK Karlsruhe, they showed this method for integrating bulk TiNi actuators onto polymer microvalves . 3.2.2 Electrical interconnection One of the advantages of heterogeneous integration is that the electrical interconnections between the integrated devices can be made very short and dense, reducing parasitic inﬂuences. There are applications, where electrical interconnection between the integrated devices is not necessary (such as the integration of SMA, which is described in later sections). Yet, if electrical interconnections are necessary (as for example in the micromirror arrays), these interconnections are in most cases electrically conductive vertical vias. The methods for providing electrical vias can be distinguished in the via ﬁrst and the via last approach. In the via ﬁrst approach, all vertical electrical interconnections are fabricated prior to integration and during bonding the devices are electrically interconnected as illustrated in ﬁgure 12a. One of the advantages of this approach is that the two substrates can be fabricated completely separately, are then electrically interconnected during the integration and no further post-integration processing is required for the electrical contacting. However, this method requires a careful alignment of the two substrates, which limits the size reduction of the vias. In the via last approach, only the electrical contact pads on the target substrate are fabricated prior to integration. After the bonding, vias are etched into the integrated substrate and ﬁlled with conductive material, as illustrated in ﬁgure 12b. In this approach, no precise wafer-to-wafer alignment is necessary and the vias can be made considerably smaller than in the via ﬁrst approach. However, the integrated substrate 22 Wafer-level heterogeneous integration of MEMS actuators electrical contacts substrate removed preparing electrical contacts the devices are electrically prior to integration interconnected during the bonding (a) via first electrical contacts preparing electrical contacts only on target devices substrate removed bonding of the substrates vias etching of vias in the integrated devices and filling them with conductive material (b) via last Figure 12. Illustration of the two approaches for electrical via fabrication: (a) the via ﬁrst approach, where the connections are fabricated prior to integration and (b) the via last approach, where the connections are fabricated after integration. must be processed after the bonding. As an example for the via last approach, the silicon micromirrors in the mirror arrays (ﬁgure 9b) are electrically connected by vias etched through the silicon and the sacriﬁcial layer to the electrical contact pads on the IC substrate. These vias are ﬁlled with metal to connect the mirrors to the driving circuitry and allow for electrostatic actuation of the mirrors. 3.2.3 Wafer-bonding techniques A key technology of heterogeneous integration is the bonding of wafers to each other. The following brief descriptions are based on wafer bonding review papers [68,69]. In principle, all the mentioned bonding methods are suitable for heterogeneous integration. In solder bonding [70, 71, 72], layers of metal or metal-alloy based solders are used to bond two wafers. The metal layers are usually deposited on both wafers, which are joined and heated to the melting temperature of the solder. The solder reﬂows and wets both wafer surfaces, causing intimate contact and bonding of the surfaces. Example solder materials are lead-tin (Pb–Sn), gold-tin (Au–Sn) and tincopper (Sn–Cu) solders. Oxides at the metal surfaces can result in poor bonding and therefore most solder bonding processes use ﬂux to remove the oxides. To some extent, solder bonding tolerates particles and structures at the wafer surfaces. The method provides hermetic bonding/packaging. Furthermore, it allows for combined bonding and vertical electrical interconnection, making it very interesting for heterogeneous integration. Eutectic bonding [73,74,75,76,77] is a variation of solder bonding, allowing to join 3 HETEROGENEOUS INTEGRATION 23 two wafers with dissimilar surface materials which form a eutectic mixture at temperatures much lower than their melting temperatures. The most common material combination is silicon (Si) and gold (Au) with a eutectic temperature of 363 °C. Eutectic bonding can result in strong and hermetic bonds at relatively low temperatures and is therefore often used for the hermetic sealing of micromachined transducers. Furthermore, the method is interesting for heterogeneous integration because it allows for vertical electrical interconnection. In adhesive bonding [69, 78, 79, 80], an intermediate adhesive layer creates a bond between two surfaces. Most commonly, a polymer adhesive is applied and the wafers are pressed together. Then, the polymer adhesive is hard-cured, typically by exposing to heat or ultraviolet (UV) light. The main advantages include the relatively low bonding temperatures between room temperature and 450 °C (depending on the polymer material), the insensitivity (to some extent) to the topology or particles on the wafer surfaces, the compatibility with standard complementary metal-oxide (CMOS) semiconductor wafers and the ability to join practically any wafer materials. While adhesive wafer bonding is a comparably simple, robust and low-cost process, concerns such as limited temperature stability and limited data about the longterm stability of many polymer adhesives in demanding environments need to be considered. Also, adhesive wafer bonding does not provide hermetically sealed bonds towards gasses and moisture. The method does not provide electrical interconnection. In direct or fusion bonding [81,82,83], two wafers are contacted without signiﬁcant pressure, electrical ﬁelds or intermediate layers. For reliable bonding, this method requires very ﬂat and very clean wafer surfaces, room temperature contacting of the wafers and an annealing step (typically between 600 and 1200 °C) to increase the bond strength. This method results in strong and hermetic bonds and is therefore of interest if the integration method should also provide hermetic packaging. The method does not provide electrical interconnection. Anodic or ﬁeld assisted bonding [84,85] is based on joining an electron conducting material such as silicon and a material with ion conductivity such as alkali-containing glass. Heating to temperatures of 180–500 °C mobilizes the ions and an applied voltage of 200–1500 V creates a large electric ﬁeld that pulls the wafer surfaces into intimate contact and fuses them together. Anodic bonding is more tolerant to surface roughness than direct bonding and usually leads to strong and hermetic bonds. The method is interesting if hermetic packaging is required, however, the large voltages might damage IC devices on the substrates. The method does not provide electrical interconnection. Thermocompression bonding, metal-to-metal direct bonding, and ultrasonic bonding [86, 87, 88, 89] are related bonding schemes in which two surfaces are pressed together and heated. Typically at least one of the surfaces consists of a metal. The surfaces plastically deform and fuse together. Instead of heating, the energy can also be supplied by ultrasonic energy (ultrasonic bonding), with the advantage of breaking through native oxides, particles and surface nonuniformities at the bond interface. Common bonding surface materials are gold to gold, copper to copper, aluminum to gold, and aluminum to glass. The disadvantage of thermocompression and ultrasonic bonding is that large net forces are required when bonding larger wafer areas. Thus, thermocompression bonding, metal-to-metal direct bonding and ultrasonic bonding 24 Wafer-level heterogeneous integration of MEMS actuators are mainly used in wire bonding schemes and in bump bonding schemes. However, these methods provide hermetic bonding/packaging and the possibility for vertical electrical interconnection, making it very interesting for heterogeneous integration. In low-temperature melting glass bonding  an inorganic low-temperature melting glass or glass frit layer forms the intermediate bonding material and is deposited on one or both of the wafers. The wafers are joined and heated, causing the glass to deform or reﬂow and bonding the wafers. Two diﬀerent types of glasses are available; devitrifying glasses, of which the melting point is permanently increased after the curing and vitreous glasses, which always melt at the same temperature. This method allows to hermetically bond various wafer materials at relatively low bonding temperatures and tolerates to some extent particles and structures at the wafer surfaces. The method does not provide electrical interconnection. 3.2.4 Releasing structures for actuation When integrating actuators there are some more issues to consider beside wafer bonding and vertical electrical interconnection. Actuators imply moving structures which must be detached from their underlying bonding layer to allow their movement. For structures fabricated using wafer-to-wafer bonding, the techniques to detach them from their underlying substrate can be summarized in two approaches. The ﬁrst method is the localized bonding of areas to be aﬃxed while avoiding the bonding of the structures to be detached. The second method is a bond-and-release approach, in which all structures are bonded to the substrate, followed by removing the bond interface material underneath the structures to be detached. Localized bonding (illustrated in ﬁgure 13a) between two substrates can be obtained using two diﬀerent principles. The ﬁrst principle is to modify the interface material prior to bonding, deﬁning bonding and non-bonding areas. Examples of patterned bond interface layers include adhesive layers applied only on areas where bonding is desired  and bond blocking layers such as gold or platinum deﬁning local non-bonding areas in anodic bonding . The second localized bonding principle is to use heat triggered bonding methods and to localize the heat to the desired areas of the bond interface. Examples of this approach include integrated heaters for both localized eutectic and silicon fusion bonding , localized soldering using inductive heating  as well as local heating using lasers . In localized bonding, the non-bonded parts are either fallout-structures or they must remain mechanically connected to the bonded parts by mechanical supports to prevent them from falling out during the remaining process steps. The removal of mechanical support structures through dicing or through controlled fracture has been shown [95, 96]. However, such break-away structures limit the design freedom and potentially increase the footprint area of the MEMS device. Furthermore, the moving structures could be damaged while removing the support structures. The most common technique for releasing bonded structures is the bond-andrelease approach based on sacriﬁcial underetching (illustrated in ﬁgure 13b). This technique requires the fabrication of the structures on top of a ’sacriﬁcial’ layer, which can be etched with a high selectivity. This approach is common in surface 3 HETEROGENEOUS INTEGRATION to remain bonded 25 to be detached not bonded bonded (a) localized bonding of the structures to remain bonded to be detached, with etch holes to remain bonded, without etch holes underetch distance w w/2 bonding layer (b) sacrificial underetching with etch holes to be detached to remain bonded bonding layer bonding layer is locally dissolved bonding layer is not attacked (c) localized removal of the bonding layer Figure 13. Illustration of the diﬀerent methods to detach structures from the substrate: (a) localized bonding of areas to be aﬃxed while avoiding the bonding of the structures to be detached; (b) sacriﬁcial underetching with etch holes; (c) localized removal of the bonding layer only underneath the structure to be released. micromachining , where the layers are stacked upon each other. In the crossbar switches presented later in this thesis, the actuators are fabricated using surface micromachining and released prior to integration with the target substrate. A wide variety of sacriﬁcial materials have been demonstrated, such as silicon dioxide , polymers [99, 100] and metals . In structures fabricated by wafer-bonding, sacriﬁcial underetching is based on wafer-bonding methods with intermediate bonding layers that can be sacriﬁcially etched with a high selectivity to release the attached structures. Examples of such intermediate bonding materials are silicon dioxide  and polymers [69, 103]. It should be noted that processes using the buried oxide in silicon-on-insulator (SOI) wafers as sacriﬁcial layer can be considered as such a bond-and-release technology. A challenge in sacriﬁcial underetching is to provide the selectivity between structures to be detached and structures to remain bonded. If the whole sacriﬁcial layer is removed, there is no selectivity and all structures are detached. One solution for this issue is to aﬃx the structures not to be detached via a second clamping material, which is not attacked when etching the sacriﬁcial layer. An example for this strategy are again the silicon micromirror arrays: the silicon is transfer integrated using adhesive polymer bonding onto a polymer sacriﬁcial layer. Then, metal vias which provide both electrical and mechanical interconnection from the silicon to the IC substrate are fabricated (see ﬁgure 9b). Finally, all polymer is removed but the mirrors are still clamped by the metal vias, which are not attacked during the polymer ashing. Another approach which does not require a second clamping, is to provide the 26 Wafer-level heterogeneous integration of MEMS actuators selectivity by integrating vertical etch holes in the structures to be detached but not in the substrates to remain bonded. These etch holes drastically minimize the distance to underetch as compared to structures without etch holes. Thus, the structures with etch holes are detached while the structures without etch holes are only partially underetched at their edges (see ﬁgure 13b). Etch holes are commonly included in sacriﬁcial underetching to minimize the time the whole substrate is exposed to the etchant or the etch process environment, which may result in device destruction if other materials than the sacriﬁcial layer are attacked. Yet, beside the advantages there are also some challenges to consider. Etch holes potentially decrease the mechanical stability and the performance of the structures to be released. Furthermore, the fabrication of such etch holes is feasible only for structures consisting of thin layers, such as the thin silicon layers of micromirror arrays. If the structures are hundreds of micrometers thick, etch holes are diﬃcult to fabricate with the required aspect ratio. Even if such deep and thin vertical etch holes could be fabricated, the etching of the underlying layers would presumably be very slow due to poor mass transport and depletion of the etchant in the etch holes. An alternative, which allows for the area-independent release of bonded structures without additional etch holes and harsh chemical environments, is the localized removal of the intermediate bonding layers (illustrated in ﬁgure 13c). An example of localized removal is the localized laser ablation of a polymer bonding layer . Another example is based on the electrochemical etching of metal sacriﬁcial layers in a neutral salt solution. This principle has been shown for detaching surface micromachined structures , which were deposited onto aluminum as sacriﬁcial layer. In this thesis, this method was adapted to release eutectically bonded silicon microstructures to allow for their manipulation by heterogeneously integrated SMA actuators. Further information about this release etch technique can be found in the attached paper 4. 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS 4 27 Knife gate microvalves with bulk SMA microactuators This chapter introduces methods for wafer-level integrating the shape memory alloy Titanium-Nickel (TiNi) and their combination with knife gate microvalves which are a new tool for controlling large gas ﬂows. First experiments proof the great promise of the knife gate valve concept, however, the thermal bimorph actuator of the ﬁrst test devices proved not robust enough and bulk TiNi material was identiﬁed as a promising replacement. There were no reported methods to wafer-level integrate bulk TiNi actuators with silicon structures and, thus, no methods for combining bulk TiNi with the knife gate microvalve concept. TiNi is diﬃcult to integrate with conventional methods and therefore large parts of this thesis deal with developing methods for heterogeneous integration of bulk TiNi material with silicon structures and methods for cold-state reset and concepts for thermal energy supply. The following sections will brieﬂy introduce the challenges of integrating TiNi material with silicon structures and the concept of knife gate microvalves. Finally, ﬁrst concepts to integrate bulk TiNi sheets and wires for actuating the valves will be presented, together with initial measurements indicating an outstanding performance of the TiNi actuated knife gate microvalves. 4.1 Integration of Titanium-Nickel shape memory alloy Despite the advantages, the shape memory alloy Titanium-Nickel (TiNi) is not yet a standard MEMS actuator material, partially due to the diﬃculties to integrate TiNi with silicon structures. So far, TiNi is integrated either monolithically or in a hybrid fashion. In monolithic integration, the TiNi is deposited directly onto the silicon structures using metal deposition techniques such as evaporation or sputtering. After deposition, the ﬁlm is amorphous, i.e. it displays no shape memory eﬀect (SME) and is compressively stressed. To functionalize the layer, the ﬁlm is crystallized in an annealing step, which involves heating the ﬁlm to typically above 535 °C . The annealing can even be performed by heating the substrate during deposition [106, 13]. During crystallization the stress in the ﬁlm relaxes and after subsequent cooling the ﬁlm is under tensile stress, which can be utilized to form a bimorph actuator. In the cold state, the silicon structure (typically a cantilever) is stiﬀ enough to overcome the tensile stress and remains ﬂat, thereby deforming the TiNi and providing the cold-state 28 Wafer-level heterogeneous integration of MEMS actuators (a) (b) Figure 14. The thin SMA ﬁlm actuated microgripper, consisting of two TiNi/silicon bimorphs which are eutectically bonded together: (a) a schematic illustration  and (b) a photograph of the device . reset. Upon heating, the TiNi ﬁlm contracts and forces the silicon cantilever to bend. Upon cooling the silicon cantilever ﬂattens the bimorph again. A well known example for this technique is the microgripper [21, 107], which is shown in ﬁgure 14. In the bimorph approach, the cold-state reset is provided by the built-in stress between the TiNi and the silicon, thus, the cold-state reset is integrated monolithically with the TiNi. Another approach is to remove the silicon underneath the TiNi, which results in a free standing TiNi ﬁlm. This method requires additional elements to provide the cold-state reset. Commonly, such additional elements include a bias spring and a mechanical spacer, with the spacer placed in between the bias spring and the TiNi ﬁlm. During the assembly of the parts, the bias spring deﬂects the TiNi via the spacer. When heated, the TiNi ﬂattens and deﬂects the bias spring via the spacer. These elements are integrated either on wafer-level by bonding of other wafers containing the cold-state reset elements or on device-level in a hybrid fashion. Figure 15 shows examples for both cases [108, 1]. Monolithic integration comes with several advantages, such as the wafer-level integration of the material and the possibility of a built-in cold-state reset when using the bimorph approach. However, monolithic integration of TiNi also comes with some major challenges . The transformation temperature of TiNi is very sensitive to variations of Ti or Ni content. An increase in the atomic percentage of nickel by 1% from an equi-atomic ratio of Ti:Ni lowers the martensitic start temperature from 50 °C to -100 °C. Therefore the deposition technique must provide precise and reliable control of the deposition rate of the two diﬀerent metals, which is diﬃcult to achieve. Even if a ﬁlm has a proper global composition, local inhomogeneities result in the nucleation and growth of precipitates that inhibit shape memory behavior. The purity of the TiNi targets and high deposition vacuum are essential to limit impurities in the deposited ﬁlm, which also potentially inhibit the SME. The high temperature annealing step causes compatibility problems with other materials and requires additional layers to avoid unwanted interdiﬀusion between TiNi and silicon. Also, the large 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS (a) 29 (b) Figure 15. Schematic drawings of TiNi actuated microvalves based on the free-standing TiNi ﬁlm approach with (a) silicon bias spring and spacer integrated via wafer-bonding  and (b) external, pick-and-place integrated metal bias spring and saphire ball spacer . thermal stresses after deposition and during annealing could cause delamination or cracking of the ﬁlms. Furthermore, the thicknesses of the ﬁlms are limited to approximately 20 μm  and a recent report states that TiNi-based ﬁlm sputtering is mostly feasible for thicknesses up to 10 μm only , potentially resulting in limited mechanical robustness of structures actuated by TiNi ﬁlms. For completeness it should be mentioned that in a recent report  a 30 μm thick ﬁlm of a ternary TiNi based alloy was achieved by ﬂash-evaporating TiNiCu onto copper substrates, however, the issues of complicated processing and high post-deposition annealing temperatures are still not overcome. One method to address these issues is the hybrid integration, where the TiNi is fabricated separately in dedicated factories and pick-and-place integrated as bulk material on device level. Bulk TiNi is commercially available in diﬀerent shapes and dimensions, for example wires with diameters starting from 18 μm and thin sheets starting from 20 μm (even thicknesses down to 5 μm have been reported ). Similar to the free-standing thin TiNi ﬁlm approach, the cold-state reset is provided by additional elements, which are pick-and-place integrated on device level. As an example, ﬁgure 16 shows a schematic drawing [111, 5] illustrating the principle of hybrid TiNi integration for microvalves. Hybrid integration of bulk TiNi comes with several advantages. The material is produced with standard metalworking techniques in dedicated factories, resulting in speciﬁed and reproducible material properties and thereby addressing the sensitivity to variations of Ti and Ni content in the alloy. However, the pick-and-place integration of the TiNi in combination with the pick-and-place integration of the cold-state reset elements is diﬃcult for microdevices and potentially results in large fabrication costs for larger volume production. 30 Wafer-level heterogeneous integration of MEMS actuators Figure 16. Schematic illustration of hybrid integration of TiNi to fabricate a microvalve [111, 5]. The bulk TiNi element is patterned and pick-and-place integrated with the other parts, including the spacer for the cold-state reset. An applied pressure pushes the membrane up, which via the spacer deforms the TiNi element. When heating, the TiNi pushes the membrane down to the ﬂow oriﬁce, thereby closing the valve. Heterogeneous integration methods allow for the combination of the two methods by integrating bulk TiNi material and providing a cold-state reset on wafer-level. There is one report on heterogeneous TiNi integration , in which a SMA sheet was patterned on wafer-scale and the elements were selectively transferred to single plastic microvalves using the selective transfer integration approach of IBM  mentioned above. However, the cold-state reset was provided by a spacer between the microvalve and the SMA element, which again requires pick and place assembly. Large parts of this thesis deal with wafer-level integration of both TiNi and the cold-state reset. 4.2 4.2.1 Gas microvalves Background A valve is a device that regulates the ﬂow of gas or liquids. In this thesis, the focus lies on millimeter sized valves for gas ﬂow regulation. It would go beyond the scope of this thesis to review all the research eﬀorts into microvalves, therefore the interested reader is referred to review papers [112, 113] for a more detailed overview. Most of the active microvalves use mechanical moving parts which are coupled to integrated actuators. Depending on the ﬂow control principle, these microvalves can be divided into diaphragm/seat type valves and gate type valves. Diaphragm/seat type valves comprise a boss (most common is a ﬂexible diaphragm) which moves in parallel to the gas ﬂow (see ﬁgure 17a) and which is coupled to an actuator. In the closed state, the boss completely closes the oriﬁce and the actuator must overcome the pneumatic pressure, thus, . a strong actuator is necessary for controlling high pressure. To allow for a large ﬂow in the open state, the distance and thereby the ﬂow channel between the seat and the boss should be maximized, thus, the actuator must provide a large stroke . KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS Boss 31 Guide Guide Guide 4 Gate Guide orifice orifice gas flow gas flow (a) (b) Figure 17. Illustration of: (a) seat valve with the ﬂow regulating boss moving in parallel to the gas ﬂow, requiring a strong and large-stroke actuator; (b) gate valve with the ﬂow regulating gate moving perpendicular to the ﬂow, reducing the requirements on the actuator. In contrast to seat valves, the gas ﬂow in gate valves is regulated by a gate which moves perpendicular to the gas ﬂow (see ﬁgure 17b). In the ideal case, when fully opening the valve the gate is completely retracted out of the ﬂow channel and there is no obstruction in the ﬂow path, resulting in higher ﬂow rates compared to a similar sized diaphragm/seat type valve. When moving the gate to close the ﬂow path, the pneumatic pressure is counteracted by the mechanical guidance of the gate and in the ideal case the actuator is only moving the gate and must not counteract the pneumatic pressure. However, the limited actuation energy available in microsystems does not allow friction between sliding structures and therefore requires spacing between ﬂow oriﬁce and gate, which results in leak ﬂow in the closed valve state. Fortunately, many valve applications tolerate leak ﬂow. Previously published work in the ﬁeld of gate microvalves always featured gates that move in the wafer plane [115, 116, 117, 118, 119]. Figure 18  shows schematic drawings of a sliding plate gate valve , which is currently commercialized by the company Microstaq [120, 121, 122]. Yet, to allow the gate to move in the wafer-plane requires extra footprint area and increases the footprint-related costs. The silicon footprint consumed by the valve is a good cost indicator and when comparing microvalves to each other not only the performance in terms of controlled ﬂow and pressure should be considered, but also the performance obtained per consumed footprint area. Therefore, in this thesis the comparison ﬁgure ’controlled ﬂow/pressure drop/footprint’ is suggested with the unit [sccm/kPa/mm2]. Table 4 shows the performance per footprint of diﬀerent microvalves and the numbers indicate, that gate microvalves typically have a much larger performance than diaphragm/seat microvalves. Microvalves should be fabricated cost-eﬃciently to allow them to be successful outside the speciﬁc niche markets they currently focus on. This thesis presents the knife gate microvalve concept, in which the gate moves out of the wafer plane to minimize the chip footprint area and the footprint-related cost. 32 Wafer-level heterogeneous integration of MEMS actuators Figure 18. Schematic drawings of the sliding gate microvalve introduced by Williams et al , showing (a) a cross-section and (b) a three-dimensional drawing of the three silicon wafers, which are assembled to form a pressure-balanced microvalve. The ﬁgure is copied from . 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS 33 Table 4. Comparison of diﬀerent valves in terms pneumatic performance relative to their consumed footprint area. It should be noted, that not all of the listed valves are silicon valves, however, they are all microvalves. The performance comparison is related to the knife gate valve presented in this work. related to Reference and actuation method Q P w l Q/P/A 3400 95 2.3 2.3 4.21 valve in the attached paper 1 this work, knife gate microvalves thermal bimorph, attached paper 1, externally actuated 100% SMA (TiNi sheet) actuation 1000 8 6.5 6.5 2.95 70% SMA (TiNi wire) actuation 3800 50 2.0 4.5 8.44 200% other gate microvalves , thermal 6700 1000 2.5 5.0 0.54 13% 5000 100 4.0 4.0 3.13 74% 500 40 1.4 1.4.0 6.17 147% 5000 140 6.0 6.0 0.99 23.59% 1000 690 8.0 5.0 0.04 0.86% 1150 800 4.0 4.9 0.09 2.14% 100 170 2.5 2.5 0.09 2.24% 2000 200 10.0 10.0 0.10 2.38% Yang et al , electrostatic 45 900 1.64 1.64 0.02 0.44% Li et al , PZT 39 210 20.0 20.0 0.00 0.01% Yang et al , PZT 52 2070 8.4 5.0 0.00 0.01% Shao et al , PZT 70 50 13.0 13.0 0.01 0.20% 800 550 3.6 3.6 0.11 2.67% Barth et al , thermal bimorph 1000 344 8.8 8.8 0.04 0.89% Yanget al , thermopneumatic 1000 228 8.0 8.0 0.07 1.63% 3 16 3.0 8.0 0.01 0.19% Kohl et al [135, 5], SMA (TiNi) 1600 120 20.0 11.0 0.06 1.44% Kohl et al [136, 5], SMA (TiNi) 470 500 6.0 6.0 0.03 0.74% Kohl et al [137, 5], SMA (TiNi) 180 210 6.0 11.0 0.01 0.31% Kohl et al [137, 5], SMA (TiNiPd) 360 250 6.0 11.0 0.02 0.52% Kohl et al [138, 5], SMA (TiNiCu) 360 300 6.0 11.0 0.02 0.43% Kohl , SMA (TiNi) 280 200 6.0 6.0 0.04 0.92% Kohl et al [139, 5], SMA (TiNi) 460 800 6.0 6.0 0.02 0.38% , thermal , externally magnetic seat microvalves Zdeblick et al , phase change, (’FLUISTOR’ valve) TiNi Alloy company, SMA (TiNi), ﬁgure 15,  Messner et al , thermal bimorph Jerman , thermal bimorph Fu et al , magnetic Huﬀ et al , electrostatic Bosch et al , magnetic + electrostatic Q = ﬂow in sccm; P = pressure in kPa; w,l = width, length in mm; A = w×l = footprint in mm2 Q/P/A = performance per footprint in sccm/kPa/mm2 34 Wafer-level heterogeneous integration of MEMS actuators ga actuator gas channel gas channel gas channel actuator actuator s gate gas (a) front gate valve gate ga s gate (b) side gate valve (c) back gate valve Figure 19. Illustration of the three diﬀerent knife gate microvalve designs introduced in detail in the attached paper 1. 4.2.2 Knife gate microvalves This thesis presents the knife gate microvalve concept, in which the gate moves out of the wafer plane to minimize the chip footprint area and the footprint-related costs. Figure 19 illustrates the three diﬀerent designs of microvalves with the gate moving out-of-plane. The gate is coupled to a monolithically integrated thermal silicon/aluminum bimorph actuator and moves out-of-plane to regulate an in-plane ﬂow. The valve concept is explained in detail in the attached paper 1. Figure 20 conﬁrms the valve’s potential for controlling large ﬂows: an active chip area of only 2.3×3.7 mm2 allows for a ﬂow change of ΔQ = 3400 standard cubic centimeters per minute (sccm) at a pressure diﬀerence of 95 kPa. Table 4 lists the performance per footprint of diﬀerent microvalves. For simpler comparison the presented knife gate microvalve is chosen as reference and all the other microvalves are compared to it. The comparison shows that the knife gate microvalve provides the highest performance per footprint, only outperformed by a recently published rotary gate valve and the TiNi wire actuated knife gate microvalve. The rotary valve is electromagnetically actuated using an external magnetic ﬁeld and belongs therefore to another class of microvalves with external actuation mechanisms. 5000 100% open flow [sccm] 4000 50% open 3000 2000 closed 1000 0 20 40 60 pressure [kPa] 80 leak flow from imperfect assembly leak flow from gate spacing Figure 20. Measured pressure-ﬂow characteristics of a back-gate microvalve. The dashed line indicates the expected leak caused by imperfect assembly. 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS 35 polymer bonding layer TiNi sheet actuator gate base gas flow carrier substrate eutectic bonding layer (a) locally remove bonding layer (b) (c) Figure 21. Cross-sectional drawing of the ﬁrst design of a TiNi sheet actuated knife gate valve showing (a) a back gate valve bonded to a carrier substrate and the TiNi sheet prior to integration; (b) the adhesive bonding of the TiNi sheet onto the valve and (c) releasing the ﬂow regulating gate to allow for its manipulation by the actuator. The TiNi wire actuated knife gate microvalve is a successor of the thermal bimorph knife gate microvalve and will be discussed below. However, it should be noted that while the experiments indicate the excellent capabilities of the knife gate microvalve concept, the thermal bimorph actuator failed for several reasons. First, the bimorph beam was not strong enough to withstand the vibration and torsional forces caused by the ﬂow. Second, improper thermal design caused improper heating and actuation of the beam. Bulk TiNi actuators were identiﬁed as a promising replacement, however, concepts for integrating bulk TiNi materials had to be developed ﬁrst. 4.3 4.3.1 TiNi sheet actuated knife gate valves Design In a ﬁrst, yet unpublished design to integrate TiNi sheets onto silicon knife gate microvalves, the TiNi is integrated with a back gate valve (back gate valve see ﬁgure 19c). Figure 21 shows the design, in which the back-gate valve is bonded onto a carrier substrate with a vertically through-etched hole, providing an enclosed ﬂow channel which is expected to simplify the pneumatic connection. The TiNi is bonded onto the valve: one end of the TiNi is bonded to the ﬂow channel forming the base for the cantilever and the other end is bonded onto the ﬂow-regulating gate. 4.3.2 Fabrication - Integration of TiNi sheets A method was investigated to adhesively bond TiNi sheets onto silicon structures, together with methods for the cold-state reset and thermal energy supply. The attached paper 2 contains a detailed description of these methods. The concept to integrate TiNi sheets onto silicon structures is illustrated in ﬁgure 22a. The TiNi sheet is patterned to form the actuator structures, using ﬂexure interconnections to hold them together. The patterning is performed prior to integration to avoid the exposure of the silicon target structures to the very aggressive TiNi etchant. The bonding adhesive is stamped onto the surface of the patterned target 36 Wafer-level heterogeneous integration of MEMS actuators wafer- sized, bulk TiNi sheet patterning of TiNi sheet flexures holding cantilevers together aligning to target wafer target wafer coated with adhesive layer bonding to target wafer etched holes to allow for cantilever movement (a) TiNi sheet integration concept cold state stress layer hot state resistive heater bulk TiNi (b) trimorph actuator concept Figure 22. Illustration of (a) ﬁnal bulk TiNi sheet integration concept and (b) trimorph actuator concept, as developed in paper Nr. 2. wafer, which was a ’dummy’ wafer with etched holes to allow the TiNi actuators to deﬂect freely. The stamping method is specially developed in this thesis and based on standard dicing blue tape. Then, the TiNi sheet is aligned and bonded onto the target wafer. For developing this integration method, the target wafer was a ’dummy’ wafer with etched holes, which allow the TiNi actuators to deﬂect freely. After the TiNi material is integrated, the cold-state reset and thermal energy supply must be provided for actuation. In this work, these elements are integrated monolithically with the TiNi sheet in form of three functional layers. Figure 22b illustrates the actuation concept. The ﬁrst functional layer is the core actuation layer formed by bulk TiNi material. The second functional layer is a stressed ﬁlm which deforms the actuator in the cold-state. This ﬁlm is deposited onto the TiNi and eliminates the need for pick-and-place integration of an additional bias spring. The third functional layer is a heating layer to supply thermal energy to the actuator via an indirect heating scheme and to avoid the complicated electrical contacting of the TiNi alloy. In the valve design shown in ﬁgure 21, the TiNi sheet is bonded onto the gate, which itself is bonded to another silicon wafer. To allow the gate to be manipulated for controlling the ﬂow, its underlying bonding layer must be removed without removing the bonding layers underneath the other silicon structures. Section 3.2.4 presents some methods for the localized release of structures while keeping other structures bonded, 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS flow [sccm] 1800 37 open state 1400 1000 600 closed state 200 0 5 10 pressure [kPa] 15 Figure 23. First measurements of the pressure-ﬂow characteristics of a yet unpublished TiNi-sheet actuated knife gate microvalve. however, most of them require underetching of the structures to be released, including exposure of the whole device to harsh environments and etching of all bonding layers. To avoid these issues, a new bond-and-release method was developed: The valve and the carrier substrate are bonded together using eutectic Au-Si bonding, followed by localized removal of the eutectic bonding layer. For a detailed description and investigation of this method, the reader is referred to the attached paper 4. 4.3.3 Results Figure 23 shows ﬁrst results of the ﬂow performance of a gate microvalve with TiNi sheet actuation. An active chip area of only 6.5×6.5 mm2 allows for a ﬂow change of ΔQ ≈ 1000 sccm at a pressure diﬀerence of 8 kPa. These numbers result in a ﬂow performance per footprint (see Table 4) of about 70% of the ﬁrst knife gate valve with the non-functional thermal bimorph actuators. However, in contrast to the ﬁrst knife gate valve, the TiNi sheet actuated valve was actuated with the integrated actuator and did not require any external actuation. Furthermore, the tested TiNi sheet knife gate valve is an early prototype with room for optimization. 4.3.4 Discussion The TiNi sheets are integrated directly onto the silicon structures, without any intermediate carrier or transfer substrate. Since the sheet is patterned prior to integration, it has to be aligned to the target wafer. Both alignment and placing onto the adhesive layer were performed manually. However, the patterned TiNi sheet was fragile and very easily deformable, which complicated the handling. The alignment of the sheet was enabled only by a high ambient temperature caused by a hotplate underneath the target wafer, which triggered the stiﬀ hot state of the sheet. To allow for an alignment and bonding using a standard wafer alignment and bonding tool, future designs should aim for a transfer integration method, in which the SMA sheet is preprocessed on an intermediate carrier substrate and then transfer bonded from the carrier substrate onto the target wafer. 38 Wafer-level heterogeneous integration of MEMS actuators probe needles TiNi cantilever silicon eutectic Au-Si bonding area heater TiNi cantilever with stress layer (a) (b) Figure 24. (a) Photographs of a eutectically bonded TiNi cantilever which was severely plastically deformed by manual pulling with a tweezer without breaking the bond, thereby illustrating the extreme bond strength. (b) Picture of a eutectically bonded TiNi trimorph actuator with a resistive heater. The sheets are aﬃxed onto the silicon structures using adhesive polymer bonding. The adhesion strength looks promising, however, more tests are necessary to verify the bond stability, especially for long-term cycling of the actuator. An alternative method could be the one used for the micromirror arrays presented in section 3: the adhesive bonding could serve as temporary mechanical ﬁxation and electroplated metal vias through the TiNi sheets could provide both electrical interconnection and permanent mechanical ﬁxation. Yet another alternative is to use a diﬀerent bonding method which allows for direct mechanical ﬁxation. Figure 24a shows non-published pictures of ﬁrst experiments testing the eutectic bonding of TiNi to silicon and illustrating the extreme bond strength: a eutectically bonded TiNi cantilever was severely plastically deformed by manual pulling with a tweezer without breaking the bond. At this stage, an electrical interconnection between the TiNi and the silicon structures is not necessary, since any scheme for electrical contacting of the actuator would most likely involve wire bonding directly to the metal heater on top of the TiNi trimorph actuators. Since the work in the attached paper 2 was focused on integration and cold-state reset, the test structures did not contain a metal heater. Figure 24b shows a photograph of a yet unpublished complete trimorph actuator containing a resistive metal heater, which is eutectically bonded to silicon. The heater was contacted using probe needles and consumed approximately 100 mW to trigger the TiNi hot state. 4.4 4.4.1 TiNi wire actuated knife gate valves Concept Figure 25 shows a ﬁrst, yet unpublished design to integrate TiNi wires with the knife gate valve concept. The wires are integrated with a front gate valve (front gate valve see ﬁgure 19c). 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS 39 polymer bonding TiNi wire gate gas flow (a) (b) (c) Figure 25. Cross-sectional drawing of the ﬁrst design of a TiNi wire actuated knife gate valve showing (a) a front gate valve and the TiNi wires prior to integration; (b) the adhesive bonding of the TiNi wires onto the valve and (c) the actuation. 4.4.2 Fabrication - Integration of TiNi wires For the SMA wire actuators, the integration concept is derived from the actuation concept, which is illustrated in ﬁgure 26a. TiNi wires are strained by pulling them prior to integration. Then, the strained wires are integrated with silicon cantilevers by anchoring one end of the wires to the base of the silicon cantilever and the other end of the wire to the moveable end of the cantilever. Upon heating, the TiNi wires recover the strain, i.e. they contract and thereby force the silicon cantilever to bend out of plane. In the cold state, the silicon cantilevers act as cold-state reset, recovering their ﬂat shape by stretching the TiNi wires. Figure 26b illustrates the resulting wafer-level integration concept. The wires are oriented and placed on a dedicated tensioning frame, which is used to strain the wires. Then, the wafer and the wires are aligned to each other and put in contact. Finally, the wires are anchored for each actuator and the tensioning frame is removed. After integration, the wires are diced together with the silicon chips. For a detailed investigation including the proof of concept by fabricating ﬁrst test structures the reader is referred to the attached paper 3. The fabricated test devices showed strokes of up to 396 μm for a cantilever length of 3 mm, which is among the highest values in comparison to diﬀerent microactuators reported in a recent review paper . Since the work in this paper was focused on integration and cold-state reset, no heating concept is presented. 4.4.3 Results Figure 27 shows ﬁrst results of the ﬂow performance of a knife gate microvalve with TiNi wire actuation. An active chip area of only 2×4.5 mm2 allows for a ﬂow change of ΔQ ≈ 3800 sccm at a pressure change of 50 kPa. These numbers result in a ﬂow performance per footprint (see Table 4) of about 200% of the ﬁrst knife gate valve with non-functional thermal bimorph actuators. In contrast to the ﬁrst knife gate valve, the TiNi wire valve is actuated with the integrated actuator. The ﬂow performance per consumed footprint area is a world record compared to the other valves and clearly demonstrates the promise of TiNi actuated knife gate microvalves. 4.4.4 Discussion The TiNi wires are transfer-integrated, using a dedicated tensioning frame as temporary carrier. For the ﬁrst test devices, the alignment was performed manually and the wires were placed onto the adhesive anchors using a dedicated lift stage. This process 40 Wafer-level heterogeneous integration of MEMS actuators operational states of the actuator TiNi wire anchor strained TiNi wire anchor silicon straining wire cold state hot state (a) actuator concept fixed clamp polymer anchors moveable clamp TiNi wires target wafer with silicon cantilevers clamping wires straining wires aligning and contacting anchoring (b) wafer-level integration concept Figure 26. Illustration of (a) the actuation concept using TiNi wires on silicon cantilevers and (b) concept for wafer-level integration of the wires, as developed in the attached paper 3. open state flow [sccm] 6000 4000 closed state leakage due to imperfect assembly 2000 0 0 10 20 30 40 pressure [kPa] 50 theoretical flow lines Figure 27. First measurements of the pressure-ﬂow characteristics of a yet unpublished TiNi-wire actuated knife gate microvalve. 4 KNIFE GATE MICROVALVES WITH BULK SMA MICROACTUATORS electrical connector pins 41 TO 8 housing wirebonding valve pneumatic connector tube Figure 28. Illustration of the packaging concept for the knife gate microvalves. The valve is packaged in a standard TO8-housing, like it is used for pressure sensors. The Picture shows a pressure sensor of the Robert Bosch GmbH with the lid removed. could be adapted to implement a high precision (x,y,z) stage, thus, the process looks promising for the wafer-level integration of wires. However, prior to integration the wires need to be placed onto the tensioning frame with the correct orientation and pitch, which was performed manually in this work. Future work should verify that the wires can be placed onto the tensioning frame using automated methods. The wires are attached to the silicon cantilevers using anchors made of SU-8, which is locally cured with UV light. However, the area underneath the wires is shadowed by the wires and therefore not fully cured. Future work should implement reﬂective patterns in the silicon underneath the wires to reﬂect the UV light into the SU-8 volume underneath the wires. Alternatively, the wires could be anchored using metal clamping via electroplating metallic connections between the silicon and the wires. 4.5 Outlook The work presented above is a big step towards a microvalve with potential for commercial application. However, a fully integrated TiNi actuated valve with integrated heaters has not been shown so far. All necessary elements have been discussed and future work will focus on combining all elements to show a fully functional TiNi actuated microvalve. For commercialization the valve must be packaged. Figure 28 shows the suggested packaging concept, which is inspired by the packaging of several commercial pressure sensors. The valve is glued onto the pneumatic connector tube of a TO8-housing, thereby pneumatically interconnecting the valve to an inlet pressure. The outlet of the ﬂow can be provided by a hole in the lid of the TO8-housing. The electrical interconnection to the actuator on the valve is provided by wirebonding to the electrical interconnection pins of the TO8-housing. 42 Wafer-level heterogeneous integration of MEMS actuators 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES 5 43 Automated main distributing frames with S-shaped actuator switches This chapter presents a MEMS crossbar switch as the core unit to automate main distributing frames, in which the parties in copper-wire telecommunication networks are cross-connected. The crossbar switch consists of an array of MEMS switches which are individually addressable and allow to interconnect a number of input lines to a number of output lines in any combination desired. In this work, the switches are based on the electrostatic S-shaped actuator described earlier in section 2.3.2. Each of the switches in the array is individually addressable using a row/column addressing method, avoiding a complicated CMOS circuitry as utilized in micromirror arrays. The fabrication concept of the crossbar switch is based on the basic concept of heterogeneous integration, which involves the separate fabrication of parts followed by their assembly to a device. The following subsections brieﬂy introduce automated main distributing frames, followed by a description of the S-shaped actuator switch. Finally, the MEMS crossbar switch and the concept of individual switch addressing are presented. 5.1 Switch units in automated main distributing frames The current telecommunication networks evolved from the original telephone networks and they are for the most part still based on copper-wire networks. Within these networks, the main distributing frame (MDF) is the center of cross-connecting equipment of service providers to their customers and it is located both in large central oﬃces and remote street cabinets. The MDF still relies upon the same routines and equipment it did 50 years ago. Hence, whenever a cross-connection must be changed, this re-conﬁguration must be carried out manually by technicians physically changing the interconnection wires (jumpering) in the MDF. Before the deregulation of the telecommunication markets the customers only rarely changed their services and there were no or little economical incentives to automate the jumpering. Nowadays, however, the customers are highly ﬂexible and there is a growing need to adapt quickly to new telecommunication services. Due to the increased labor-intensive manual jumpering the network maintaining costs are drastically rising and thereby creating a clear incentive to automate the jumpering using an automated MDF (AMDF). The core element of the AMDF is a crossbar switch. Originally, the term ”crossbar 44 Wafer-level heterogeneous integration of MEMS actuators + + + +++output channels (a) electrical diagram of a switch unit input channels double switch output channels input channels crossbar switch unit (b) multicast switching networks Figure 29. Electrical and hierarchical schematic diagrams: (a) switch unit, consisting of an array of 3 × 3 double switches; (b) arrangement of switch units on higher hierarchical layers using the Clos method to minimize the necessary amount of switches. switch” relates to a grid of metal bars crossing each other without touching each other and which are electrically interconnected or isolated by switches in their crosspoints. In a AMDF, the metal bars are formed by signal channels, with a set of input channels and a set of output channels. A channel consists of a signal line pair and switching the signal line pair simultaneously requires a double-pole single-throw (DPST) switch in each cross-point. Typically, the cross-bar switches are symmetrical, with N inputs and N outputs resulting in N2 crosspoints. Figure 29a shows an electrical schematic of the signal circuit for a simpliﬁed crossbar switch with N=3. A typical AMDF connects tens of thousands of channels and it is not feasible to provide this amount of crosspoints in a single crossbar switch since it is diﬃcult to fabricate such large arrays with an acceptable yield. Furthermore, each bar or channel can only be connected in one crosspoint while the other crosspoints remain unused and therefore large parts of the crossbar switch are not utilized. A solution is to fabricate small crossbar switch units with a low N and interconnect several of them to a bigger crossbar switch unit, which itself again forms a switch unit in a bigger crossbar switch. In 1953, Charles Clos at Bell labs developed the ’minimal spanning switch’ method  to interconnect the small crossbar switch units in a way to reduce the total number of switches and interconnections, but still allowing for a non-blocking network which always provides one route to interconnect an input to an output. In this method, the switch units are arranged in three stages with each the same number of switch units and each switch unit sharing only one connection with a switch unit in another stage, as illustrated in ﬁgure 29b. If a path is blocked, the connections can be rerouted via other switch units to open a new path. When scaling up the resulting crossbar switch, Clos’ interconnection method reduces the√ amount √ of √ crosspoints from the quadratically scaling N2 to the more favorable 3× N × N × N = 3×N1.5 , thereby drastically minimizing the number of switches and the related costs. As an example, a conventional crossbar switch with N = 100 would require N2 = 10000 crosspoint switches, whereas a Clos crossbar switch requires 10 switch units à 10×10 crosspoints 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES 45 Figure 30. Illustration of the switch unit in a robot based AMDF: The signal channels are interconnected in their crosspoints by inserting/extracting metal pins using a robot on a xy-stage. The ﬁgure is copied from . in three stages, which results in 10×10×10×3 = 3000 crosspoint switches, only. It would go beyond the scope of this thesis to extend this introduction, therefore the interested reader is referred to literature such as [140, 141, 142, 143, 144]. A crossbar switch unit comprises three basic elements. The ﬁrst element is the routing network with input and output signal channels crossing each other without touching each other. The second element are switches needed for interconnecting the signal channels in their crosspoints. The third element is controlling of the switches: each switch must be individually addressable to allow for the interconnection of any of the N input lines to any of the N output lines in any conﬁguration desired. To ensure the highest ﬂexibility when connecting new services and new technological equipment to the customers, the signal path should ideally act like a copper-wire cable which requires ohmic mechanical switching with metal contacts and excludes solid-state relay based crossbar switching. Previous attempts to provide a switch unit for a Clos network in an automated MDF consist of robots [145, 146, 147] or electromagnetic relays [148, 149]. Figure 30  exemplarily illustrates the robot approach: the routing network is formed by a matrix board and the switches are metal pins, which are inserted/extracted in the crosspoints of the signal channels using a robot. The addressing is provided by moving the robot to the addressed crosspoint using a xy-stage with a laser for ﬁne alignment. In the electromagnetic relay approach, the crossbar switching is performed by an array of electromagnetic relays on printed circuit boards. These arrays include control circuits with diodes and transistors to allow for a row/column addressing. The attached paper 5 contains a short comparison of the technologies, which did not provide solutions with the desired performance at acceptable costs Nowadays, MEMS technology oﬀers the possibility to integrate a large number of micro-switches on a single chip and utilizes high-volume semiconductor manufactur- 46 Wafer-level heterogeneous integration of MEMS actuators top electrode stiff cantilever bottom electrode switching contact large gap between electrodes (a) OFF (b) ON Figure 31. Illustration of a conventional cantilever-type electrostatic MEMS switch. The ﬁgure is modiﬁed from . ing methods and logistics which potentially results in low production costs. MEMS switches have been investigated extensively during the last decade for many applications requiring high quality signal switching [150, 151, 152]. They ﬁll the gap between solid-state relays (SSR) and electromagnetic relays (EMR) by oﬀering true ohmic switching, high miniaturization and integration and very good signal properties over a large bandwidth. To the authors’ knowledge, there are no scientiﬁc reports on MEMS crossbar switches or switch arrays for the described application. However, there are a few reports on small-scaled and unpackaged metal-contact MEMS switch arrays for other applications [153, 154, 155, 156, 157, 158]. 5.2 5.2.1 MEMS switches Background MEMS switches are devices that mechanically open or short-circuit a transmission line. They are of submillimeter size and they outperform semiconductor switches in terms of power consumption, on-state resistance, cut-oﬀ frequency, isolation and signal insertion loss [150,151,152]. There are two basic types of MEMS switches: capacitive switches and metal-contact switches. Capacitive switches are typically used as shunt switches between the signal line and the ground line while metal-contact switches are used as series switches between two transmission lines, opening or closing the signal path. In metal-contact switches, the signal is conducted through metal contacts and in capacitive switches, the contact is made between the metal and dielectric material. The application in automated MDF’s described above requires ohmic switching of signals from DC to max. 50 MHz, therefore only metal contact series switches are of interest in this work. For metal contact switches there are several requirements to consider when designing the switch. In the oﬀ-state, the switch should provide a high isolation when applying signals with high voltages, hence, a large distance between the switch contacts is necessary. When the switch is closed, the contact resistance should be as low as possible and the contact should remain stable during the whole ON time. The switch contact resistance and reliability is directly inﬂuenced by the contact force and the contact material. The inﬂuence of the contact materials is a research ﬁeld on its own [159, 160, 161, 162] and not touched upon in this work. The contact force should be maximized, however, a strong contact force increases the risk of stiction between 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES 47 Figure 32. Illustration of the connection between the design parameters and the active forces in the switch model and how they aﬀect its performance in terms of contact resistance and reliability. The ﬁgure is copied from . the contacts. Therefore, the forces opening the switch should be strong enough to overcome the adhesion forces between the switch contacts. Most of the MEMS switches are electrostatically actuated and the conventional and so far most common electrostatically actuated MEMS switch is based on a simple cantilever approach, as illustrated in ﬁgure 31. The switch contact bar at the moveable end of the cantilever interconnects two signal lines when the cantilever is pulled down, as illustrated in ﬁgure 31b. This conﬁguration is basically a parallel-plate actuator and comes with all the problems mentioned in section 2.3.1, especially in combination with the switch requirements described above: a large distance between the switch contacts for a high OFF state signal isolation results in a large initial electrode gap and thereby in large actuation voltages. To overcome stiction forces when opening the switch, a stiﬀ cantilever is required to provide strong opening forces. As a consequence, increased actuation voltages are necessary to overcome the strong spring force. Figure 32  illustrates all the parameters involved for electrostatically actuated metal contact switches and how they are interconnected. 5.2.2 The S-shaped actuator switch To overcome the limitations of the conventional cantilever type switch, an alternative approach was developed based on the S-shaped ﬁlm actuator described in section 2.3.2 [35, 36, 37]. The assembly concept of the S-shaped actuator highlights this actuator for the heterogeneous integration with other structures. One substrate contains all the moveable parts of the actuator, while the target substrate provides an actuation electrode and a spacer to provide the gap in which the membrane moves up and down. Figure 33 illustrates the basic concept of the S-shaped ﬁlm actuator 48 Wafer-level heterogeneous integration of MEMS actuators switch substrate switch contact signal lines to connect d V target substrate (a) before assembly (b) after assembly: OFF (c) ON Figure 33. The S-shaped actuator switch: (a) before assembly, (b) after assembly in the OFF-state and (c) in the ON-state. switch. The switch substrate contains the ﬂexible membrane with a switch contact and the target substrate contains the signal lines which are to be interconnected by the switch (ﬁgure 33a). The distance d between the switch contacts in the OFF state (ﬁgure 33b) can be tuned and a large distance allows to obtain a high oﬀ-state isolation. In the ON state (ﬁgure 33c), the signal lines are interconnected by the switch contact on the membrane. The switch contact area is surrounded by the actuation electrode, resulting in a more homogeneous contact force. If the spring force of the thin rolling membrane is not strong enough to open the switch, the upper zipper-actuator provides an active opening capability. 5.3 The MEMS crossbar switch The following subsections introduce the MEMS crossbar switches developed in this thesis. These crossbar switches are fully fabricated using standard MEMS technologies and materials. Thus, strictly speaking no heterogeneous integration is applied since heterogeneous integration refers to the integration of diﬀerent technologies or materials. However, the development of the MEMS crossbar switch is based on the basic concept of heterogeneous integration; the device is divided into substructures, which are fabricated separately and ﬁnally integrated with each other. For the MEMS crossbar switch, the elements to integrate are a signal channel routing network, switches to interconnect the signal channels in their crosspoints and a control line network for the individual addressing of single switches. The following sections brieﬂy introduce the routing network, the switches and their integration to a crossbar switch. Finally the addressing concept will be presented. A detailed description of the concept and the fabrication can be found in the attached paper 5 and 6. 5.3.1 The routing network Figure 34 (bottom part) illustrates the signal line routing network. The signal lines are fabricated in two diﬀerent metal layers, which are isolated from each other by an intermediate polymer layer. The input lines are in the lower metal layer and the output lines are in the upper metal layer, crossing the input lines. In each crosspoint the input lines are connected to the upper metal layer by vertical metal vias through the 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES top part with switches to interconnect signal lines in their crosspoints double switch in the crosspoints assembly 49 membrane electrode top electrode spacer ring and adhesive layer assembly intermediate isolation layer sign s elec ignal al trod + e input on lower metal layer al + sign nal sig vertical via through intermediate isolation layer crosspoint of input and output output + bottom electrode on upper metal layer bottom part with signal line routing network Figure 34. Three-dimensional drawing showing the concept of the KTH MEMS crossbar switch, simpliﬁed to a 2 × 2 array. Some components of the bottom part have been removed to reveal details. The bottom part contains the signal-routing network on two metal layers and the bottom electrodes for the switch actuation. Furthermore, the picture shows the two BCB layers for isolating the metal layers of the routing network from each other and for creating the encapsulating spacer ring between the top and the bottom part. The top part contains the moving part of the S-shaped ﬁlm actuator switches with the switch contacts to interconnect the signal lines in their crosspoints. 50 Wafer-level heterogeneous integration of MEMS actuators intermediate polymer layer. The upper metal layer also contains actuation electrodes (bottom electrodes) to actuate the switches later on. 5.3.2 The crosspoint switches Figure 34 (top part) illustrates the switching part of the crossbar switch, which allows to interconnect the signal lines in each crosspoint. The switches are S-shaped actuator switches as introduced above. As illustrated in the electrical schematic of the signal circuit in ﬁgure 29a, the signal lines are paired to channels, consisting of a signal+ and a signal− line. To switch both lines simultaneously, each membrane contains two separate switch contacts to allow for the parallel switching of two line pairs in the target substrate. 5.3.3 The integration Finally, the top part with the switches is placed onto the bottom part with the signal channels. The spacing between the two parts is provided by a polymer ring (ﬁgure 34, spacer ring) around the crosspoints. Besides spacing, the polymer ring is also designed to provide the mechanical ﬁxation of the two parts. The polymer ring is patterned prior to the assembly, but not fully cured. After the assembly, the polymer is cured and thereby the two parts are adhesively bonded. Furthermore, the switches are encapsulated/packaged by the polymer ring, based on the encapsulation/packaging concept illustrated in ﬁgure 10. 5.3.4 Individual switch addressing There are several possibilities for the individual addressing of the switches. The conventional approach would be to connect each switch with an individual control wire. However, the number of switches and thereby the number of control wires scale quadratically with the size of the matrix, making this conventional approach feasible only for very small arrays with a small number of elements. Another approach is to integrate a CMOS addressing circuitry as shown earlier for micromirror arrays, which allows for the analog control of the actuators in the array as well as for high addressing frequencies . The presented fabrication concept of the MEMS crossbar switch would allow for the heterogeneous integration of the switches with a CMOS wafer, however, the AMDF application requires only digital mode addressing with low addressing frequencies and does not justify the integration of high-cost CMOS circuits. The addressing approach developed for the MEMS crossbar switch utilizes the hysteresis behavior of the actuation voltages of electrostatic actuators, which was explained in section 2.3.1 and illustrated in ﬁgure 3c. Applying a set voltage Vset which is larger than the pull-in voltage always pulls in the moveable plate and the switch is ON: Vset ≥ Vpull−in . Reducing the applied voltage to a hold voltage Vhold within the hysteresis range between the pull-in and the pull-out voltage holds the moveable plate in the current position, i.e. either in the ON or in the OFF state: 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES 51 -Vset- Vhold Vhold Vhold Vpull-out < Vhold < Vpull-in Vhold Vhold Vhold + Vset+ Vhold + Vset+ - (-Vset-) > Vpull-in Vhold - (-Vset-) < Vpull-in Vhold Vhold Vhold Vpull-out < Vhold < Vpull-in Vhold + Vset+ < Vpull-in Vpull-out < Vhold < Vpull-in (a) (b) (c) Figure 35. Schematic drawings illustrating the individual setting of cell (2,1) in a 3×3 array and the voltages involved. (a) Hold state. Each actuator maintains its current state. (b) Pull-in of the actuator by applying the addressing voltages at the respective rows and columns. (c) Return to hold state. Maintaining the programmed states of all cells. Vpull−out < Vhold < Vpull−in . To open the switch, the applied voltage is reduced to the reset voltage Vreset , which is smaller than the pull-out voltage: Vreset ≤ Vpull−out . For the addressing scheme to work, the electrodes of the ﬁxed plates are interconnected in each respective column, whereas the electrodes of the moveable plates are interconnected in each respective row. This arrangement allows to provide the required potential diﬀerences in the crosspoints of the electrodes and the number of necessary control lines is reduced from the quadratically scaling N2 to the linear scaling complexity of 2N. As an example, with the conventional addressing approach a crossbar switch with N = 20 would require 400 control lines, while the presented row/column approach requires only 2N = 40 control wires. The individual pull-in is illustrated in ﬁgure 35. If all columns are fed with the hold voltage Vhold and all rows are on ground potential, all switches maintain their current states (ﬁgure 35a). To set the actuator at the address (2,1), the potential diﬀerence between the electrodes must exceed the pull-in voltage Vpull−in . Therefore, a voltage level of Vhold + Vset+ is applied at the 2nd row, and a negative voltage level of Vset− is applied at the 3rd column (ﬁgure 35b). After the individual pull-in the voltages ΔVset+ and ΔVset− are removed and the columns are connected to the hold voltage Vhold , maintaining the current state of all switches as illustrated in ﬁgure 35c. The individual pull-out is illustrated in ﬁgure 36. All columns are fed with the hold voltage Vhold and all rows are on ground potential, all switches maintain their current states (ﬁgure 36a). To reset the actuator at the address (2,1), the potential diﬀerence between the electrodes must be below the pull-out voltage Vpull−out . Therefore, a positive voltage level of Vreset is applied at the 3rd column and thereby subtracted 52 Wafer-level heterogeneous integration of MEMS actuators Vreset Vhold Vhold Vhold Vpull-out < Vhold < Vpull-in Vhold Vhold Vhold + + Vreset Vreset Vhold - Vreset < Vpull-out Vhold Vhold Vhold Vpull-out < Vhold < Vpull-in Vpull-out < Vhold < Vpull-in Vhold + Vreset - Vreset = Vhold Vhold +Vreset < Vpull-in (a) (b) (c) Figure 36. Schematic drawings illustrating the individual resetting of cell (2,1) in a 3×3 array and the voltages involved. (a) Hold state. Each actuator maintains its current state. (b) Pull-out of the actuator by applying the addressing voltages at the respective rows and columns. (c) Return to hold state. Maintaining the programmed states of all cells. from Vhold . To avoid resetting all the switches in the 3rd column, Vreset must be added to Vhold at the other rows (ﬁgure 36b). Finally, all columns are connected to the hold voltage Vhold , maintaining the current state of all switches as illustrated in ﬁgure 36c. Another simpler method for the resetting is to connect all electrodes to ground and set a new pattern as described above. This addressing scheme has been reported before for micromirror arrays [163,164, 165,166,167] and programmable capacitor banks . Neither of these reports, however, discusses any design requirements or considerations on the operational robustness of the addressing scheme. This thesis contains a full investigation for determining the operational reliability of the addressing scheme and to optimize the addressing voltage levels, taking into account the statistical deviations of the actuation voltages both for a single actuator as well as across the array. The attached journal paper 6 extensively investigates this issue and presents methods to choose actuation voltages which result in a statistical certainty of better than 99.9% for the addressing to work. 5.4 Discussion The routing network and the switches are integrated directly, without any intermediate carrier or transfer substrate. The mechanical ﬁxation of the two substrates is designed as adhesive bonding via the patterned adhesive spacer ring. In this design, the polymer spacer ring not only provides spacing and mechanical ﬁxation, but also an encapsulation and thereby packaging of the crosspoints. However, it should be noted here, that even though the mechanical ﬁxation of the two substrates is designed as a wafer-level process, the ﬁrst test structures were assembled manually on 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES 53 Figure 37. Picture of the electrical contacting to evaluate the crossbar switch. The arrow indicates the probe needle from the bottom side to contact the electrodes on the top substrate of the crossbar switch. chip-level using epoxy droplets in the corners for mechanical ﬁxation. This issue will be addressed in more detail below. Usually, a controlled atmosphere inside the packaged volume is desired and therefore hermetic sealing of the encapsulation would be the optimum. The polymer based sealing of the presented crossbar switches does not provide hermetic sealing even if the chosen polymer, Bencocyclobutene (BCB) from the Dow Chemical Company, is often referred to as ”near-hermetic” packaging . A full hermetic packaging is obtained using a metallic bonding scheme, however, at the cost of the relatively noncomplicated method to provide the spacing between the substrates using a polymer ring. The two substrates interact with each other both electrically and mechanically. Yet, there are no vertical electrical interconnection vias which allow to access the switch control electrodes on the switch substrate via the network substrate. The lack of this vias is the main reason, why the ﬁrst test structures were assembled manually on chip-level: both the switch substrate (top part) and the network substrate (bottom part) contain switch electrodes with contact pads. For the assembly the switch substrate is ﬂipped over and placed onto the spacer on the network substrate. Now the electrode contact pads of the switch substrate cannot be accessed from the top anymore and so far there are no vertical electrical interconnection vias allowing to link the electrodes on the switch substrate to contact pads on the network substrate. The chosen contacting method is illustrated in ﬁgure 37: the device is assembled on chip-level with an overlap between the rims of the two substrates and the contact pads are accessed from the top and from the bottom using probe needles. For the early prototypes, this method was suﬃcient to evaluate the feasibility of the crossbar switches, but future devices should contain vertical electrical interconnections. 54 Wafer-level heterogeneous integration of MEMS actuators Figure 38. Picture showing a MEMS crossbar switch after the assembly. A weight was necessary to force down the top part, counteracting the spring forces of the membranes, as illustrated in the left-hand inset. Then, the top part with the switches was aligned to the bottom part and ﬁnally the two substrates were ﬁxated with epoxy droplets in the corners. The moveable parts of the switches are released prior to integration: the membranes are fabricated by plasma enhanced chemical vapor deposition (PECVD) of silicon nitride on a polymer sacriﬁcial layer. Assisted by etch holes in the membrane, the polymer was removed in a plasma ashing step. As a consequence, the thin membranes are not mechanically supported during the integration of the switch substrate, potentially damaging the membranes. However, the manual assembly has shown that the membranes of the switches are surprisingly mechanically robust. As an example: during the manual assembly, the top part with 400 silicon nitride membranes with a thickness of 1 μm was placed on the bottom part. Because of large stress gradients in the membranes the deﬂection of the free ends were much larger than the thickness of the spacer ring and the top part had to be forced down onto the spacer ring using a weight as shown in ﬁgure 38. In this state, the switches were manually aligned (alignment accuracy ±5 μm), with considerable friction between the tips of the strained membranes and the surface of the bottom part. Despite the friction and the resulting strains and stresses, in most cases all the membranes survived the procedure. The stress gradients in the membranes are high, resulting in high actuation voltages especially for the 20×20 array (∼ 95 V) with shorter membranes and smaller electrode area. Therefore the PECVD process should be optimized to allow for more control over the stress gradients in the deposited ﬁlms. Furthermore, when depositing the metal ﬁlm for the membrane electrode additional stresses are induced between the metal and the membrane. Also, the devices got very hot during the removal of the sacriﬁcial layer underneath the membranes in an oxygen plasma . The combination of heat and plasma inﬂuences the stresses in the silicon nitride membranes [170,171,172], which issues should be further investigated. Most of the switches failed to provide a good ohmic contact. Assumed reasons are 5 AUTOMATED MAIN DISTRIBUTING FRAMES WITH S-SHAPED ACTUATOR SWITCHES 55 that the contact force was not high enough and the surfaces of the switch contacts were contaminated since the switch was not assembled in a cleanroom environment. In future devices, the contact force can be increased by lowering the stress-gradient in the membrane and by local stiﬀening of the membrane to improve the transfer of the electrostatic force between the surrounding electrodes to the switch contacts. To avoid contamination, future devices should be assembled in the cleanroom, ideally on wafer-level and after a short plasma cleaning of the surfaces. The actuators were very susceptible to stiction, which was most probably caused by electrostatic stiction between the electrodes. In general, electrostatic stiction is an issue for electrostatic actuators where dielectric layers are involved. Charges are injected and trapped in the dielectric layers, thereby causing an electrostatic attraction. In the MEMS crossbar switch, charges can be trapped in the silicon nitride of the membranes and the electrical isolation layers as well as in the large amount of BCB underneath the bottom electrodes. Consequently, in future devices the amount of dielectric materials should be kept at a minimum. 5.5 Outlook Currently a second generation of MEMS crossbar switches is under development, which features arrays of 20×20 double-switches on a silicon footprint of only 12×12 mm2 . In the revised fabrication no complicated electroplating of thick metal layers is necessary and the overall stress gradients in the membranes can be controlled using a stress-compensation design. The chip is fully hermetically packaged by metallic waferbonding and contains vertical electrical interconnection between the two substrates as well as through-silicon vias through the network substrate to allow for hybrid integration as introduced in section 3. To obtain a more reliable operation of the switches all polymers are removed to reduce the risk of electrostatic stiction caused by trapped charges. 56 Wafer-level heterogeneous integration of MEMS actuators 6 6 SUMMARIES OF THE APPENDED PAPERS 57 Summaries of the Appended Papers Paper 1: Out of plane knife gate microvalves for controlling large gas ﬂow This paper considers design issues for microvalves for large gas ﬂow control. It introduces out-of-plane knife-gate microvalves as a novel design concept and a proportional microvalve concept for pressure control applications. The design of three diﬀerent actuator-gate conﬁgurations and ﬁrst prototypes are presented. The ﬁrst valve prototypes feature thermal silicon–aluminum bimorph actuators and the pressure-ﬂow performance per chip area of the demonstrator valve presented is greatly increased using out-of-plane actuation and an out-of-plane oriﬁce. The characterization of the actuators and of the pressure-ﬂow performance is presented. Paper 2: Wafer-scale manufacturing of bulk shape memory alloy microactuators based on adhesive bonding of Titanium-Nickel sheets to structured silicon wafers This paper presents a concept for the wafer-scale manufacturing of microactuators based on the adhesive bonding of bulk shape memory alloy (SMA) sheets to silicon microstructures. Wafer-scale integration of a cold-state deformation mechanism is provided by the deposition of stressed ﬁlms onto the SMA sheet. A concept for heating of the SMA by Joule heating through a resistive heater layer is presented. Critical fabrication issues were investigated, including the coldstate deformation, the bonding scheme and related stresses and the TitaniumNickel (TiNi) sheet patterning. Novel methods for the transfer stamping of adhesive and for the handling of the thin TiNi sheets were developed, based on the use of standard dicing blue tape. First demonstrator TiNi cantilevers, waferlevel adhesively bonded on a microstructured silicon substrate, were successfully fabricated and evaluated. Intrinsically stressed silicon dioxide and silicon nitride were deposited to deform the cantilevers in the cold state and the resulting tip deﬂections were evaluated. Paper 3: Design and wafer-level fabrication of SMA wire microactuators on silicon This paper reports on the fabrication of micro actuators through wafer-level integration of pre-strained SMA wires to silicon structures. In contrast to previous work, the wires are strained under pure tension, and the cold state reset is provided by single crystalline silicon cantilevers. The fabrication is based on standard MEMS manufacturing technologies and enables an actuation scheme featuring high work densities. A mathematical model is discussed, which 58 Wafer-level heterogeneous integration of MEMS actuators provides a useful approximation for practical designs and allows analyzing the actuators performance. Prototypes have been tested and the inﬂuence of constructive variations on the actuator behavior is theoretically and experimentally evaluated. The test results are in close agreement with the calculated values, and show that the actuators feature displacements among the highest reported. Paper 4: Localized removal of the Au-Si eutectic bonding layer for the selective release of microstructures This paper presents and investigates a novel technique for the footprint and thickness-independent selective release of Au–Si eutectically bonded microstructures through the localized removal of their eutectic bond interface. The technique is based on the electrochemical removal of the gold in the eutectic layer and the selectivity is provided by patterning the eutectic layer and by proper electrical connection or isolation of the areas to be etched or removed, respectively. The gold removal results in a porous silicon layer, acting similar to standard etch holes in a subsequent sacriﬁcial release etching. The paper presents the principle and the design requirements of the technique. First test devices were fabricated and the method successfully demonstrated. Furthermore, the paper investigates the release mechanism and the eﬀects of diﬀerent gold layouts on both the eutectic bonding and the release procedure. Paper 5: Single-chip MEMS 5×5 and 20×20 double-pole single-throw switch arrays for automating telecommunication networks This paper reports on microelectromechanical (MEMS) switch arrays with 5×5 and 20×20 double-pole single-throw (DPST) switches embedded and packaged on a single chip, which are intended for automating main distribution frames in copper-wire telecommunication networks. Whenever a customer requests a change in his telecommunication services, the copper-wire network has to be reconﬁgured which is currently done manually by a costly physical re-routing of the connections in the main distribution frames. To reduce the costs, new methods for automating the network reconﬁguration are sought after by the network providers. The presented devices comprise 5×5 or 20×20 double switches, which allow us to interconnect any of the 5 or 20 input lines to any of the 5 or 20 output lines. The switches are based on an electrostatic S-shaped ﬁlm actuator with the switch contact on a ﬂexible membrane, moving between a top and a bottom electrode. The devices are fabricated in two parts which are designed to be assembled using selective adhesive wafer bonding, resulting in a wafer-scale package of the switch array. The on-chip routing network consists of thick metal lines for low resistance and is embedded in bencocyclobutene (BCB) polymer layers. Paper 6: Row/Column addressing scheme for large electrostatic actuator MEMS switch arrays and optimization of the operational reliability by statistical analysis This paper investigates the design and optimization of a row/column addressing scheme to individually pull in or pull out single electrostatic actuators in an N2 array, utilizing the electromechanical hysteresis behavior of electrostatic 6 SUMMARIES OF THE APPENDED PAPERS 59 actuators and eﬃciently reducing the number of necessary control lines from N2 complexity to 2N. This paper illustrates the principle of the row/column addressing scheme. Furthermore, it investigates the optimal addressing voltages to individually pull in or pull out single actuators with maximum operational reliability, determined by the statistical parameters of the pull-in and pull-out characteristics of the actuators. The investigated addressing scheme is implemented for the individual addressing of cross-connect switches in a microelectromechanical systems 20×20 switch array, which is utilized for the automated any-to-any interconnection of 20 input signal line pairs to 20 output signal line pairs. The investigated addressing scheme and the presented calculations were successfully tested on electrostatic actuators in a fabricated 20×20 array. The actuation voltages and their statistical variations were characterized for diﬀerent subarray cluster sizes. Finally, the addressing voltages were calculated and veriﬁed by tests. 60 Wafer-level heterogeneous integration of MEMS actuators 7 7 CONCLUSIONS 61 Conclusions This thesis presents methods for the wafer-level integration of TiNi and S-shaped electrostatic actuator switches to functionalize MEMS devices. The integration methods are based on concepts of heterogeneous integration. Using the developed integration methods, TiNi actuators are integrated with a new type of microvalves, drastically improving their performance. The microvalves are knife gate microvalves and introduced in this thesis. The S-shaped actuator switches are arranged in arrays and integrated with a signal line routing network, forming a MEMS crossbar switch. In more detail, the developed TiNi integration methods: – allow to wafer-level integrate bulk TiNi sheets and wires, together with mechanisms for the cold-state reset, avoiding the complicated monolithic integration and the pic-and-place hybrid integration. First actuator test structures were fabricated and evaluated, showing promising results. Heating with integrated heaters is shown for the sheet based actuators; – provide techniques for handling the TiNi prior to integration. The TiNi sheets are patterned prior to integration to avoid exposure of the target structures to the very aggressive TiNi etchant. The thin TiNi wires are oriented and strained before they are transferred to the target substrate. – are discussed, including initial tests of alternative approaches. The developed knife gate microvalves: – feature a gate moving perpendicular to the ﬂow, reducing the demands on the actuator. The gate moves out-of-plane to control an in-plane gas ﬂow, minimizing the required silicon footprint area and the related costs; – display excellent pneumatic performance in relation to the consumed silicon footprint area, however, the ﬁrst test structures required external actuation since the integrated actuators failed; 62 Wafer-level heterogeneous integration of MEMS actuators For combining knife gate microvalves and TiNi actuators, the thesis shows: – ﬁrst designs for integrating TiNi sheets and wires with silicon knife gate microvalves, including a method for the localized release of eutectically bonded, thick silicon microstructures to allow their manipulation by the integrated actuators; – ﬁrst experiments displaying outstanding pneumatic performance of the valve in relation to the consumed silicon footprint area. – a discussion and outlook on future work, including a scheme for packaging of the valve. The MEMS crossbar switch: – is intended as core switch unit for automating parts of copper-wire telecommunication networks; – consists of two parts, with one part containing signal input and output lines crossing each other and the other part containing an array of S-shaped actuator switches. After integrating the switch array on the signal lines using adhesive bonding, the switches allow to interconnect the input to the output lines in any combination desired; – features integrated encapsulation of the switch arrays; – contains up to 20×20 = 400 crosspoint switches on an area of 12×16 mm2 . First prototypes were fabricated and evaluated, prooﬁng the feasibility of the concept; – requires a method for the individual addressing of selected switches. The thesis presents the concept of row/column addressing, drastically reducing the necessary amount of control wires. The reliability of the addressing is maximized by determining the addressing voltages based on the statistical deviation of the actuation voltages across the array. – is discussed and an outlook is given, based on current developments of a second generation of MEMS crossbar switches. 7 CONCLUSIONS (a) 63 (b) (c) Figure 39. Figure illustrating my gratitude to family, friends and colleagues in three diﬀerent languages: (a) Swedish, (b) English and (c) German. Acknowledgments This work was ﬁnancially supported by Pondus Instruments AB, by Network Automation mxc AB, by Vinnova (the Swedish Agency for Innovation Systems) through the Summit and the Forska&Väx framework, and by the European Commission through the sixth framework programme funded project Q2M. As all big projects, the project ’PhD’ requires the support of so many other people. Figure 39 expresses what I want to say to you: First of all, my principal supervisor Professor Göran Stemme for allowing me to start the PhD-adventure in his group, yet always being there to limit the adventure part and giving advice when needed. I truly appreciate that your door was always open to discuss major and minor issues, not only during my ups but mainly during my downs. My supervisors Wouter van der Wijngaart and Joachim Oberhammer for showing me how the academic world and research works. Special thanks to Wouter for not only being my supervisor, but for the feeling of being welcome in the early days and for being a great friend who helped during those downs. Frank Niklaus for sorting out the worst in the integration section of this thesis and for always having an open-minded approach to things. Björn Samel for being a true friend, the great atmosphere in our ’German oﬃce’ and for giving me the assurance that I’m not the only German ﬁghting against Swedish windmills. Thank you for our friendship ! Another good friend, Niklas Sandström, for showing me the Swedish way of life and of course for all the wonderful company when sharing hotel rooms (not like it sounds) and when enjoying our ’matlåda’ together. Henrik Gradin for being a great room-mate and for the very fruitful work we did together. You’re a smart one and I hope some of the smartness diﬀused to my side of the oﬃce. Kjell Norén for his shared interest in everything with motors and wheels and for demonstrating what a real engineer is. Thanks for all your brilliant ideas and prototypes which helped so much. Petra Palmquist for proofreading this thesis without despairing over it. Special thanks also to my former professors, Peter Pokrowsky and Patrick Klär for the support in the early days of my PhD adventure and for the remaining good contact. Of course a big ’thank you’ also to all the other colleagues I had the pleasure 64 Wafer-level heterogeneous integration of MEMS actuators to work with in my projects. Sjoerd Haasl for the guidance and great spirit in my early PhD days. Prof. Manfred Kohl, Dr. Thomas Grund and Johannes Barth from FZK Karlsruhe for our fruitful work in the Q2M project and for the honor of inviting me as speaker to your seminar. Special thanks to Thomas and the ’Deutsche Eiche’ Johannes for the nice time beside those meetings and conferences. Dr. Jan Peirs and Donato Clausi from KUL Leuven for the fruitful work in the Q2M project. Donato - I am glad I met you and I hope we will remain friends. Thorbjörn Ebefors, Göran Edström, Göran Lindvall and Jochen Walter from Silex Microsystems for the close collaboration when developing the MEMS crossbar switch 2.0. Cecilia Aronsson for introducing me to the cleanroom routines and to the Swedish language. Sveinbjörg Ingvarsdottír, Jutta Müntjes and Gaspard Pardon for enduring my guidance during your master thesis and placements. Thanks to Erika Appel and Hanne Eklund for paving the way through KTHbureaucracy and enabling things. All the work was performed in a great atmosphere, for which I want to thank my colleagues from KTH-MST: Göran, Frank, Joachim, Wouter, Hans, Erika, Kjell, Niclas, Mikael, Mikael, Mikael (yes, it’s a common Swedish name), Niklas, Fredrik and Carl Fredrik, Adit, Farizah, Andreas, Martin, Kristinn, Gaspard, Umer, Zargham, Hitthesh, Henrik, Nutapong, Staﬀan. Besides work, there is also a personal life. I would like to thank all my friends who helped me during this time by either distracting me from work or cheering me up in hard times. You know who you are - Thank you ! My family - always there for me. Thank you to my parents Regina and Kurt for demonstrating the values of honesty and reliability and how to be almost perfect parents. I have not forgotten that you should be among the research funders in the ﬁrst paragraph of the acknowledgments. Thank you to my brother Andreas for being my big brother. Thank you to my niece Charlotte for the smile when I think of you. Last, but not least - the biggest thank you to my Love Juliane for loving me the way I am. Das war meine Thesis - das Spiel geht weiter ... REFERENCES 65 References  N. Maluf and K. Williams, An introduction to microelectromechanical systems engineering, 2nd ed. Boston : Artech House, 2004.  G. T. Kovacs, Micromachined Transducers Sourcebook. McGraw-Hill, 1998.  E. Thielicke and E. Obermeier, “Microactuators and their technologies,” Mechatronics, vol. 10, no. 4-5, pp. 431 – 455, 2000.  M. J. Madou, Fundamentals of Microfabrication, 2nd ed. 2002.  M. Kohl, Shape Memory Microactuators. CRC Press LLC, Springer, 2004.  D. Bell, T. Lu, N. Fleck, and S. Spearing, “MEMS actuators and sensors: Observations on their performance and selection for purpose,” Journal of Micromechanics and Microengineering, vol. 15, no. 7, pp. 153 – 164, July 2005.  S. A. Wilson, R. P.-J. Jourdain, Q. Zhang, R. A. Dorey, C. R. Bowen, M. Willander, Q. U. Wahab, M. Willander, S. M. Al-hilli, O. Nur, E. Quandt, C. Johansson, E. Pagounis, M. Kohl, J. Matovic, B. Samel, W. van der Wijngaart, E. W. Jager, D. Carlsson, Z. Djinovic, M. Wegener, C. Moldovan, E. Abad, M. Wendlandt, C. Rusu, and K. Persson, “New materials for micro-scale sensors and actuators: An engineering review,” Materials Science and Engineering R: Reports, vol. 56, no. 1-6, pp. 1 – 129, 2007.  D. P. Arnold and N. Wang, “Permanent Magnets for MEMS,” Journal of Microelectromechanical Systems, vol. 18, no. 6, pp. 1255–1266, December 2009.  J. Comtois, V. Bright, and M. Phipps, “Thermal microactuators for surfacemicromachining processes,” Proceedings of the SPIE - The International Society for Optical Engineering, vol. 2642, pp. 10 – 21, 1995.  S. Prasanna and S. Spearing, “Materials selection and design of microelectrothermal bimaterial actuators,” Journal of Microelectromechanical Systems, vol. 16, no. 2, pp. 248–259, April 2007.  D. Hodgson and J. Brown, “Using Nitinol alloys,” Shape Memory Applications, Inc., White paper, 2000. [Online]. Available: www.sma-inc.com  F. Wang and W. Buehler, “Additional unique property changes observed during TiNi transition,” Applied Physics Letters, vol. 21, no. 3, pp. 105 – 6, August 1972.  E. Wibowo and C. Kwok, “Fabrication and characterization of sputtered NiTi shape memory thin ﬁlms,” Journal of Micromechanics and Microengineering, vol. 16, no. 1, pp. 101 – 8, January 2006.  W. Buehler, J. Gilfrich, and R. Wiley, “Eﬀect of low-temperature phase changes on the mechanical properties of alloys near composition TiNi,” Journal of Applied Physics, vol. 34, no. 5, pp. 1475 – 1477, May 1963. 66 Wafer-level heterogeneous integration of MEMS actuators  K. Otsuka and C. Wayman, Shape Memory Materials. Press, 1998. Cambridge University  Z. Wei, R. Sandstrom, and S. Miyazaki, “Shape-memory materials and hybrid composites for smart systems: Part I shape-memory materials,” Journal of Materials Science, vol. 33, no. 15, pp. 3743 – 3762, 1998.  D. Stöckel, “Status and trends in shape memory technology,” in Proceedings ACTUATOR, 1992, pp. 79–84.  Y. Bellouard, “Shape memory alloys for microsystems: a review from a material research perspective,” Materials Science & Engineering: A (Structural Materials: Properties, Microstructure and Processing), vol. 481-482, pp. 582 – 9, May 2008.  H. Scherngell and A. Kneissl, “Training and stability of the intrinsic two-way shape memory eﬀect in Ni-Ti alloys,” Scripta Materialia, vol. 39, no. 2, pp. 205 – 212, 1998.  ——, “Generation, development and degradation of the intrinsic two-way shape memory eﬀect in diﬀerent alloy systems,” Acta Materialia, vol. 50, no. 2, pp. 327 – 341, 2002.  P. Krulevitch, A. Lee, P. Ramsey, J. Trevino, J. Hamilton, and M. Northrup, “Thin ﬁlm shape memory alloy microactuators,” Journal of Microelectromechanical Systems, vol. 5, no. 4, pp. 270 – 82, December 1996.  J. Van Humbeeck, “Shape memory alloys: a material and a technology,” Advanced Engineering Materials, vol. 3, no. 11, pp. 837 – 50, November 2001.  G. Eggeler, E. Hornbogen, A. Yawny, A. Heckmann, and M. Wagner, “Structural and functional fatigue of NiTi shape memory alloys,” Materials Science and Engineering A, vol. 378, pp. 24 – 33, 2004.  G. Song, “Robust position regulation of a shape memory alloy wire actuator,” Proceedings of the Institution of Mechanical Engineers. Part I: Journal of Systems and Control Engineering, vol. 216, no. 3, pp. 301 – 308, 2002.  N. Troisfontaine, P. Bidaud, and P. Dario, “Control experiments on two SMA based microactuators,” Experimental robotics V (Lecture Notes in Control and Information Sciences), vol. 232, pp. 490–499, 1998.  N. Ma, G. Song, and H.-J. Lee, “Position control of shape memory alloy actuators with internal electrical resistance feedback using neural networks,” Smart materials and structures, vol. 13, pp. 777–783, 2004.  B. Selden, K. Cho, and H. Asada, “Segmented shape memory alloy actuators using hysteresis loop control,” Smart Materials and Structures, vol. 15, pp. 642– 652, 2006. REFERENCES 67  M. Sreekumar, M. Singaperumal, T. Nagarajan, M. Zoppi, and R. Molﬁno, “Recent advanves in nonlinear control technologies for shape memory alloy actuators,” Journal of Zheijang University SCEINCE A, vol. 8, no. 5, pp. 818–829, 2007.  J. Yao, “RF MEMS from a device perspective,” Journal of Micromechanics and Microengineering, vol. 10, no. 4, pp. 9 – 38, December 2000.  E. Hung and S. Senturia, “Extending the travel range of analog-tuned electrostatic actuators,” Journal of Microelectromechanical Systems, vol. 8, no. 4, pp. 497–505, 1999.  L. Rocha, E. Cretu, and R. Wolﬀenbuttel, “Using dynamic voltage drive in a parallel-plate electrostatic actuator for full-gap travel range and positioning,” Journal of Microelectromechanical Systems, vol. 15, no. 1, pp. 69–83, 2006.  C. G. Agudelo, M. Packirisamy, G. Zhu, and L. Saydy, “Nonlinear control of an electrostatic micromirror beyond pull-in with experimental validation,” Journal of Microelectromechanical Systems, vol. 18, no. 4, pp. 914 – 923, 2009.  J. Oberhammer, A.-Q. Liu, and G. Stemme, “Time-eﬃcient quasi-static algorithm for simulation of complex single-sided clamped electrostatic actuators,” Journal of Microelectromechanical Systems, vol. 16, no. 2, pp. 373 – 382, 2007.  M. Shikida, K. Sato, and T. Harada, “Fabrication of an S-shaped microactuator,” Journal of Microelectromechanical Systems, vol. 6, no. 1, pp. 18–24, March 1997.  J. Oberhammer and G. Stemme, “Low voltage high-isolation DC-to-RF MEMS switch based on an S-shaped ﬁlm actuator,” IEEE Transactions on Electron Devices, vol. 51, no. 1, pp. 149–155, January 2004.  ——, “Design and fabrication aspects of an s-shaped ﬁlm actuator based dc to rf mems switch,” Journal of Microelectromechanical Systems, vol. 13, no. 3, pp. 421 – 428, 2004.  J. Oberhammer, “Novel RF MEMS switch and packaging concepts,” Ph.D. dissertation, KTH-Royal Institute of Technology, 2004.  O. Brand, “Microsensor integration into systems-on-chip,” Proceedings of the IEEE, vol. 94, no. 6, pp. 1160 – 76, June 2006.  G. Fedder, R. Howe, T.-J. K. Liu, and E. Quevy, “Technologies for cofabricating MEMS and electronics,” Proceedings of the IEEE, vol. 96, no. 2, pp. 306 – 22, February 2008.  J. Smith, S. Montague, J. Sniegowski, J. Murray, and P. McWhorter, “Embedded micromechanical devices for the monolithic integration of MEMS with CMOS,” International Electron Devices Meeting. Technical Digest (Cat. No.95CH35810), pp. 609 – 12, 1995. 68 Wafer-level heterogeneous integration of MEMS actuators  S. Lewis, S. Alie, T. Brosnihan, C. Core, T. Core, R. Howe, J. Geen, D. Hollocher, M. Judy, J. Memishian, K. Nunan, R. Paine, S. Sherman, B. Tsang, and B. Wachtmann, “Integrated sensor and electronics processing for > 108 "iMEMS" inertial measurement unit components,” in Electron Devices Meeting, 2003. IEDM ’03 Technical Digest. IEEE International, 2003, pp. 39.1.1–39.1.4.  J. R. Hui, W. Xingtao, and W. C., “Adressing large arrays of electrostatic actuators for adaptive optics applications,” in Proceedings of the SPIE - The International Society for Optical Engineering, vol. 5553, no. 1, October 2004, pp. 17–27.  R. Tummala, “SOP: what is it and why? A new microsystem-integration technology paradigm-Moore’s law for system integration of miniaturized convergent systems of the next decade,” IEEE Transactions on Advanced Packaging, vol. 27, no. 2, pp. 241 – 9, May 2004.  S. Al-Sarawi, D. Abbott, and P. Franzon, “A review of 3-D packaging technology,” IEEE Transactions on Components, Packaging and Manufacturing Technology, Part B: Advanced Packaging, vol. 21, no. 1, pp. 2 – 14, February 1998.  D. M. Jang, C. Ryu, K. Y. Lee, B. H. Cho, J. Kim, T. S. Oh, W. J. Lee, and J. Yu, “Development and evaluation of 3-D SIP with vertically interconnected through silicon vias (TSV),” Electronic Components and Technology Conference, pp. 847 – 52, 2007.  R. Beica, C. Sharbono, and T. Ritzdorf, “Through silicon via copper electrodeposition for 3D integration,” 58th Electronic Components and Technology Conference, pp. 577 – 83, 2008.  J. Gu, W. Pike, and W. Karl, “A novel capillary-eﬀect-based solder pump structure and its potential application for through-wafer interconnection,” Journal of Micromechanics and Microengineering, vol. 19, no. 7, pp. 074 005 (7 pp.) –, July 2009.  T. Bauer, “First high volume via process for packaging and integration of MEMS / CMOS,” Whitepaper. [Online]. Available: www.silexmicrosystems.com  S. Braun, “Praxissemesterbericht/Practical term report,” Fachhochschule Kaiserslautern/Zweibrücken, Robert Bosch GmbH, Tech. Rep., 2001.  H. Lakner, P. Durr, U. Dauderstaedt, W. Doleschal, and J. Amelung, “Design and fabrication of micromirror arrays for UV lithography,” Proceedings of the SPIE - The International Society for Optical Engineering, vol. 4561, pp. 255 – 64, 2001.  M. Lapisa, F. Zimmer, F. Niklaus, A. Gehner, and G. Stemme, “CMOS integretable piston type micro mirror arrays for adaptive optics made of mono crystalline silicon using 3D integration,” in Proceedings IEEE International Conference on Micro Electro Mechanical Systems (MEMS), January 2009, pp. 1007–1010. REFERENCES 69  J. Sampsell, “Digital micromirror device and its application to projection displays,” Journal of Vacuum Science & Technology B (Microelectronics and Nanometer Structures), vol. 12, no. 6, pp. 3242 – 6, 1994.  S. Kurth, R. Hahn, C. Kaufmann, K. Kehr, J. Mehner, U. Wollmann, W. Dotzel, and T. Gessner, “Silicon mirrors and micromirror arrays for spatial laser beam modulation,” Sensors and Actuators A (Physical), vol. A66, no. 1-3, pp. 76 – 82, April 1998.  F. Niklaus, S. Haasl, and G. Stemme, “Arrays of monocrystalline silicon micromirrors fabricated using CMOS compatible transfer bonding,” Journal of Microelectromechanical Systems, vol. 12, no. 4, pp. 465 – 9, August 2003.  T. Bakke, M. Friedrichs, B. Volker, M. Reiche, L. Leonardsson, H. Schenk, and H. Lakner, “Spatial light modulators with monocrystalline silicon micromirrors made by wafer bonding,” Proceedings of the SPIE - The International Society for Optical Engineering, vol. 5715, no. 1, pp. 69 – 79, 2004.  T. Bakke, B. Volker, D. Rudloﬀ, M. Friedrichs, H. Schenk, and H. Lakner, “Large scale, drift free monocrystalline silicon micromirror arrays made by wafer bonding,” Proceedings of the SPIE - The International Society for Optical Engineering, vol. 6114, pp. 611 402 – 1, January 2006.  “International Technology Roadmap for Semiconductors (IRTS).” [Online]. Available: www.irts.net  M. Esashi, “Wafer level packaging of MEMS,” Transactions of the Institute of Electronics, Information and Communication Engineers C, vol. J91-C, no. 11, pp. 527 – 33, November 2008.  D. Van Thourhout, G. Roelkens, J. Van Campenhout, J. Brouckaert, and R. Baets, “Technologies for on-chip optical interconnects,” 2005 IEEE LEOS Annual Meeting, pp. 204 – 5, 2005.  M. Mastrangeli, S. Abbasi, C. Varel, C. Van Hoof, J. Celis, and K. Bohringer, “Self-assembly from milli-to nanoscales: methods and applications,” Journal of Micromechanics and Microengineering, vol. 19, no. 8, pp. 083 001 (37 pp.) –, August 2009.  I. Kurniawan, M. Tichem, and U. Staufer, “Dry self-alignment for batchassembly of chips,” in Proceedings of IEEE Transducers, 2009, pp. 212–215.  K.-F. Böhringer, K. Goldberg, M. Cohn, R. Howe, and A. Pisano, “Parallel microassembly with electrostatic force ﬁelds,” Proceedings IEEE International Conference on Robotics and Automation (Cat. No.98CH36146), vol. vol.2, pp. 1204 – 11, 1998.  C. Lin, F.-G. Tseng, and C.-C. Chieng, “Orientation-speciﬁc ﬂuidic selfassembly process based on a capillary eﬀect,” Journal of Micromechanics and Microengineering, vol. 19, no. 11, pp. 115 020 (12 pp.) –, November 2009. 70 Wafer-level heterogeneous integration of MEMS actuators  U. Srinivasan, M. Helmbrecht, C. Rembe, R. Muller, and R. Howe, “Fluidic self-assembly of micromirrors onto microactuators using capillary forces,” IEEE Journal of Selected Topics in Quantum Electronics, vol. 8, no. 1, pp. 4 – 11, January 2002.  Y. A. Chapuis, A. Debray, L. Jalabert, and H. Fujita, “Alternative approach in 3D MEMS-IC integration using ﬂuidic self-assembly techniques,” Journal of Micromechanics and Microengineering, vol. 19, no. 2, pp. 105 002–105 011, 2009.  R. Guerre et al., “Selective transfer technology for microdevice distribution,” Journal of Microelectromechanical Systems, vol. 17, no. 1, pp. 157 – 165, 2008.  T. Grund, R. Guerre, M. Despont, and M. Kohl, “Transfer bonding technology for batch fabrication of SMA microactuators,” Eur. Phys. J. Special Topics, vol. 158, pp. 237–242, 2008.  M. Schmidt, “Wafer-to-wafer bonding for microstructure formation,” Proceedings of the IEEE, vol. 86, no. 8, pp. 1575 – 85, 1998.  F. Niklaus, G. Stemme, J. Q. Lu, and R. J. Gutmann, “Adhesive wafer bonding,” Journal of Applied Physics, vol. 99, no. 3, pp. 031 101 1–28, February 2006.  A. Goyal, J. Cheong, and S. Tadigadapa, “Tin-based solder bonding for MEMS fabrication and packaging applications,” Journal of Micromechanics and Microengineering, vol. 14, no. 6, pp. 819 – 25, July 2004.  C.-W. Lin, H.-A. Yang, W. C. Wang, and W. Fang, “Implementation of threedimensional SOI-MEMS wafer-level packaging using through-wafer interconnections,” Journal of Micromechanics and Microengineering, vol. 17, no. 6, pp. 1200 – 1205, 2007.  P. Robogiannakis, E. Kyriakis-Bitzaros, K. Minoglou, S. Katsafouros, A. Kostopoulos, G. Konstantinidis, and G. Halkias, “Metallic bonding methodology for heterogeneous integration of optoelectronic dies to CMOS circuits,” Microelectronic Engineering, vol. 85, no. 4, pp. 727 – 32, April 2008.  R. Wolﬀenbuttel et al., “Low-temperature silicon wafer-to-wafer bonding using gold at eutectic temperature,” Sensors and Actuators A (Physical), vol. A43, no. 1-3, pp. 223 – 9, 1994.  A.-L. Tiensuu, M. Bexell, J.-A. Schweitz, L. Smith, and S. Johansson, “Assembling three-dimensional microstructures using gold-silicon eutectic bonding,” Sensors and Actuators A (Physical), vol. A45, no. 3, pp. 227 – 36, December 1994.  R. Wolﬀenbuttel, “Low-temperature intermediate Au-Si wafer bonding; eutectic or silicide bond,” Sensors and Actuators A (Physical), vol. A62, no. 1-3, pp. 680 – 6, 1997. REFERENCES 71  S. Lani, A. Bosseboeuf, B. Belier, C. Clerc, C. Gousset, and J. Aubert, “Gold metallizations for eutectic bonding of silicon wafers,” Microsystem Technologies, vol. 12, no. 10-11, pp. 1021 – 5, Sept. 2006.  P. Chen, C. Lin, and C. Liu, “Amorphous Si/Au wafer bonding,” Applied Physics Letters, vol. 90, no. 13, pp. 132 120 – 1, March 2007.  F. Niklaus, R. Kumar, J. McMahon, J. Yu, J.-Q. Lu, T. Cale, and R. Gutmann, “Adhesive wafer bonding using partially cured benzocyclobutene for three-dimensional integration,” Journal of the Electrochemical Society, vol. 153, no. 4, pp. 291 – 5, April 2006.  F. Niklaus, A. Decharat, F. Forsberg, N. Roxhed, M. Lapisa, M. Populin, F. Zimmer, J. Lemm, and G. Stemme, “Wafer bonding with nano-imprint resists as sacriﬁcial adhesive for fabrication of silicon-on-integrated-circuit (SOIC) wafers in 3D integration of MEMS and ICs,” Sensors and Actuators: A Physical, vol. 154, no. 1, pp. 180 – 6, August 2009.  J. Oberhammer et al., “Selective wafer-level adhesive bonding with benzocyclobutene for fabrication of cavities,” Sensors and Actuators A: Physical, vol. 105, no. 3, pp. 297 – 304, 2003.  Y. Feng, “Silicon direct bonding on vacuum and force,” IEEE International Conference on Mechatronics and Automation (IEEE Cat. No. 06EX1416C), pp. 491 – 6, 2006.  S. H. Christiansen, R. Singh, and U. Gösele, “Wafer direct bonding: From advanced substrate engineering to future applications in micro/nanoelectronics,” in Proceedings of the IEEE, vol. 94, no. 12, December 2006, pp. 2060–2106.  P. Guegen, C. Ventosa, L. Di Cioccio, H. Moriceau, F. Grossi, M. Rivoire, P. Leduc, and L. Clavelier, “Physics of direct bonding: Applications to 3D heterogeneous or monolithic integration,” Microelectronic Engineering, doi:10.1016/j.mee.2009.07.030, 2009.  M. Chen, X. Yi, L. Yuan, and S. Liu, “Research on low-temperature anodic bonding using induction heating,” Journal of Physics: Conference Series, vol. 34, no. 1, pp. 973 – 8, 2006.  K. Knowles and A. T. J. van Helvoort, “Anodic bonding,” International Materials Reviews, vol. 51, no. 5, pp. 273–311, October 2006.  M. Taklo, P. Storas, K. Schjolberg-Henriksen, H. Hasting, and H. Jakobsen, “Strong, high-yield and low-temperature thermocompression silicon wafer-level bonding with gold,” Journal of Micromechanics and Microengineering, vol. 14, no. 7, pp. 884 – 90, July 2004.  G.-S. Park, Y.-K. Kim, K.-K. Paek, J.-S. Kim, J.-H. Lee, and B.-K. Ju, “Lowtemperature silicon wafer-scale thermocompression bonding using electroplated gold layers in hermetic packaging,” Electrochemical and Solid-State Letters, vol. 8, no. 12, pp. 330 – 2, December 2005. 72 Wafer-level heterogeneous integration of MEMS actuators  X. Ang, G. Zhang, B. Tan, J. Wei, Z. Chen, and C. Wong, “Direct metal to metal bonding for microsystems interconnections and integration,” Proceedings of 7th Electronics Packaging Technology Conference, pp. 390–393, 2005.  J. Kim, B. Jeong, M. Chiao, and L. Lin, “Ultrasonic bonding for MEMS sealing and packaging,” IEEE Transactions on Advanced Packaging, vol. 32, no. 2, pp. 461–467, May 2009.  R. Knechtel, “Glass frit bonding: An universal technology for wafer level encapsulation and packaging,” Microsystem Technologies, vol. 12, no. 1-2 SPEC. ISS., pp. 63 – 68, 2005.  T. Veenstra et al., “Use of selective anodic bonding to create micropump chambers with virtually no dead volume,” Journal of the Electrochemical Society, vol. 148, no. 2, pp. 68 – 72, 2001.  Y. Cheng et al., “Localized silicon fusion and eutectic bonding for MEMS fabrication and packaging,” Journal of Microelectromechanical Systems, vol. 9, no. 1, pp. 3 – 8, 2000.  H.-A. Yang, M. Wu, and W. Fang, “Localized induction heating solder bonding for wafer level MEMS packaging,” Journal of Micromechanics and Microengineering, vol. 15, no. 2, pp. 394 – 9, February 2005.  J. Kim et al., “Thin layer laser bonding using spin-on-glass materials,” Applied Surface Science, vol. 254, no. 21, pp. 6842 – 6848, 2008.  W. van der Wijngaart, A. Ridgeway, and G. Stemme, “A micromachined knife gate valve for high-ﬂow pressure regulation applications,” 12th International Conference on Solid-State Sensors, Actuators and Microsystems. Digest of Technical Papers (Cat. No.03TH8664), vol. vol.2, pp. 1931 – 4, 2003.  S. S. Nasiri et al., “Method and system of releasing a MEMS structure,” United States Patent 7250353, 2007.  J. Bustillo, R. Howe, and R. Muller, “Surface micromachining for microelectromechanical systems,” Proceedings of the IEEE, vol. 86, no. 8, pp. 1552 – 74, August 1998.  J. Buhler, F.-P. Steiner, and H. Baltes, “Silicon dioxide sacriﬁcial layer etching in surface micromachining,” Journal of Micromechanics and Microengineering, vol. 7, no. 1, pp. R1 – R13, 1997.  C. W. Storment, D. A. Borkholder, V. Westerlind, J. W. Suh, N. I. Maluf, and G. T. Kovacs, “Flexible, dry-released process for aluminum electrostatic actuators,” Journal of Microelectromechanical Systems, vol. 3, no. 3, pp. 90 – 96, 1994.  K. Walsh, J. Norville, and Y.-C. Tai, “Photoresist as a sacriﬁcial layer by dissolution in acetone,” Technical Digest. MEMS 14th IEEE International Conference on Micro Electro Mechanical Systems (Cat. No.01CH37090), pp. 114 – 17, 2001. REFERENCES 73  D. Westberg, O. Paul, G. Andersson, and H. Baltes, “Surface micromachining by sacriﬁcial aluminium etching,” Journal of Micromechanics and Microengineering, vol. 6, no. 4, pp. 376 – 84, December 1996.  O. Gigan et al., “Fabrication and characterization of resonant SOI micromechanical silicon sensors based on DRIE micromachining, freestanding release process and silicon direct bonding,” Proceedings of the SPIE - The International Society for Optical Engineering, vol. 4936, pp. 194 – 204, 2002.  N. Somjit et al., “Novel concept of microwave MEMS reconﬁgurable 7x45◦ multistage dielectric-block phase shifters,” in Proceedings of MEMS, 2009.  S. Metz et al., “Partial release and detachment of microfabricated metal and polymer structures by anodic metal dissolution,” Journal of Microelectromechanical Systems, vol. 14, no. 2, pp. 383 – 391, 2005.  S. Miyazaki and K. Nomura, “Development of perfect shape memory eﬀect in sputter-deposited Ti-Ni thin ﬁlms,” Proceedings of the IEEE Micro Electro Mechanical Systems, pp. 176 – 181, 1994.  Y. Fu, H. Du, W. Huang, S. Zhang, and M. Hu, “TiNi-based thin ﬁlms in MEMS applications: A review,” Sensors and Actuators, A: Physical, vol. 112, no. 2-3, pp. 395 – 408, 2004.  A. P. Lee, D. R. Ciarlo, P. A. Krulevitch, S. Lehew, J. Trevino, and M. A. Northrup, “A practical microgripper by ﬁne alignment, eutectic bonding and SMA actuation,” Sensors and Actuators A: Physical, vol. 54, no. 1-3, pp. 755– 759, June 1996.  C. A. Ray, C. L. Sloan, A. D. Johnson, J. D. Busch, and B. R. Petty, “A silicon based shape memory alloy microvalve,” in Mat. Res. Soc. Symp. Proc., vol. 276, 1992, pp. 161–166.  S. Miyazaki, M. Tomozawa, and H. Kim, “Development of high-speed microactuators utilizing sputter-deposited TiNi-base shape memory alloy thin ﬁlms,” in Proceedings ACTUATOR, 2008, pp. 372–377.  T. Mineta, K. Kasai, Y. Sasaki, E. Makino, T. Kawashima, and T. Shibata, “Flash-evaporated TiNiCu thick ﬁlm for shape memory alloy micro actuator,” Microelectronic Engineering, vol. 86, pp. 1274–1277, 2009.  K. Skrobanek, M. Kohl, and S. Miyazaki, “Stress-optimised shape memory microvalves,” in Micro Electro Mechanical Systems, Proceedings, IEEE., Tenth Annual International Workshop on, 1997, pp. 256–261.  P. J. Hesketh, J. S. Bintoro, and L. Luharuka, “Microvalve for fuel cells and miniature gas chromatographic system,” in Sensors Update 13, pp. 233–302.  K. W. Oh and C. H. Ahn, “A review of microvalves,” Journal of Micromechanics and Microengineering, vol. 16, no. 5, pp. R13 – R39, 2006. 74 Wafer-level heterogeneous integration of MEMS actuators  W. van der Wijngaart, A. Thorsén, and G. Stemme, “A study of oriﬁcecontrolled ﬂow for microvalve design optimisation,” IEEE Journal of Microelectromechanical Systems, vol. 14, pp. 200–206, April 2005.  K. R. Williams, N. I. Maluf, E. N. Fuller, R. J. Barron, D. P. Jaeggi, and B. P. van Drieënhuizen, “A silicon microvalve for the proportional control of ﬂuids,” in Proceedings of Transducers ’99, Sendai, Japan, 1999, pp. 1804–7.  M. Walters, V. Dhuler, R. Mahadevan, A. Cowen, R. Wood, E. Hill, and I. Kao, “A silicon micromachined gate valve,” in 1998 Solid-State Sensors and Actuators Symposium, South Carolina, USA, 1998.  A. P. Papavasiliou, D. Liepmann, and A. P. Pisano, “Fabrication of a free ﬂoating silicon gate valve,” in Proceedings of International Mechanical Engineering Congress and Exposition (IMECE), Nashville, TN, USA, 1999.  J. A. Frank and A. P. Pisano, “Low-leakage micro gate valves,” in Digest of Technical Papers, Transducers. 12th International Conference on Solid-State Sensors, Actuators and Microsystems, vol. 1, 2003, p. 143.  R. Luharuka and P. Hesketh, “A bistable electromagnetically actuated rotary gate microvalve,” Journal of Micromechanics and Microengineering, vol. 18, no. 3, pp. 035 015 – 1, 2008.  S. Booth and R. Kaine, “Fluid handling: Big gains from tiny valve,” Applicance Design, pp. 46–48, April 2008. [Online]. Available: www.applicancedesign.com  J. Uibel, “The miniaturization of ﬂow control,” in 9th International Conference on ﬂuid control, measurements, and visualization, 2007.  M. Luckevich, “MEMS microvalves: the new valve world,” Valve-world magazine, pp. 79–83, May 2007. [Online]. Available: www.valve-world.net  M. J. Zdeblick, R. Anderson, J. Jankoswki, B. Kline-Schode, L. Christel, R. Miles, and W. Weber, “Thermopneumatically actuated micro valves and integrated electro-ﬂuidic circuits,” in Technical Digest of Solid-state sensor and actuator workshop, Hilton Head Island, 1994, pp. 251–255.  S. Messner, M. Muller, V. Burger, J. Schaible, H. Sandmaier, and R. Zengerle, “A normally-closed, bimetallically actuated 3-way microvalve for pneumatic applications,” Proceedings MEMS. IEEE. Eleventh Annual International Workshop on Micro Electro Mechanical Systems. An Investigation of Micro Structures, Sensors, Actuators, Machines and Systems (Cat. No.98CH36176), pp. 40 – 44, 1998.  H. Jerman, “Electrically activated normally closed diaphragm valves,” Journal of Micromechanics and Microengineering, vol. 4, no. 4, pp. 210 – 216, December 1994. REFERENCES 75  C. Fu, Z. Rummler, and W. Schomburg, “Magnetically driven micro ball valves fabricated by multilayer adhesive ﬁlm bonding,” Journal of Micromechanics and Microengineering, vol. 13, no. 4, pp. 96 – 102, July 2003.  X. Yang, A. Holke, S. Jacobson, J. Lang, M. Schmidt, and S. Umans, “An electrostatic, on/oﬀ microvalve designed for gas fuel delivery for the MIT microengine,” Journal of Microelectromechanical Systems, vol. 13, no. 4, pp. 660 – 668, August 2004.  H. Li, D. Roberts, J. Steyn, K. Turner, O. Yaglioglu, N. Hagood, S. Spearing, and M. Schmidt, “Fabrication of a high frequency piezoelectric microvalve,” Sensors and Actuators A (Physical), vol. A111, no. 1, pp. 51 – 56, March 2004.  E.-H. Yang, C. Lee, J. Mueller, and T. George, “Leak-tight piezoelectric microvalve for high-pressure gas micropropulsion,” Journal of Microelectromechanical Systems, vol. 13, no. 5, pp. 799 – 807, October 2004.  P. Shao, Z. Rummler, and W. Schomburg, “Polymer micro piezo valve with a small dead volume,” Journal of Micromechanics and Microengineering, vol. 14, no. 2, pp. 305 – 9, February 2004.  M. A. Huﬀ, J. R. Gilbert, and M. A. Scmidt, “Flow characteristics of a pressure balanced microvalve,” in Proceeding IEEE Transducers, June 1993, pp. 98–101.  P. W. Barth, C. C. Beatty, L. A. Field, J. W. Baker, and G. B. Gordon, “A robust normally-closed silicon microvalve,” in Technical Digest of Solid-state sensor and actuator workshop, Hilton Head Island, 1994, pp. 248–250.  X. Yang, C. Grosjean, and Y.-C. Tai, “Design, fabrication, and testing of micromachined silicone rubber membrane valves,” Journal of Microelectromechanical Systems, vol. 8, no. 4, pp. 393 – 402, December 1999.  D. Bosch, B. Heimhofer, G. Muck, H. Seidel, U. Thumser, and W. Welser, “A silicon microvalve with combined electromagnetic/electrostatic actuation,” Sensors and Actuators A (Physical), vol. A37-A38, pp. 684 –692, June 1993.  M. Kohl, K. Skrobanek, and S. Miyazaki, “Development of stress-optimised shape memory microvalves,” Sensors and Actuators A (Physical), vol. A72, no. 3, pp. 243 – 50, February 1999.  M. Kohl and B. Krevet, “3D simulation of a shape memory microactuator,” Material Transactions, vol. 43, no. 5, pp. 1030–1036, 2002.  M. Kohl, D. Dittmann, E. Quandt, and B. Winzek, “Thin ﬁlm shape memory microvalves with adjustable operation temperature,” Sensors and Actuators A (Physical), vol. A83, no. 1-3, pp. 214 – 219, May 2000.  M. Kohl, D. Dittmann, E. Quandt, B. Winzek, S. Miyazaki, and D. Allen, “Shape memory microvalves based on thin ﬁlms or rolled sheets,” Materials Science and Engineering A, vol. 273, pp. 784 – 788, 1999. 76 Wafer-level heterogeneous integration of MEMS actuators  M. Kohl, J. Gobes, and B. Krevet, “Normally-closed shape memory microvalve,” International Journal of Applied Electromagnetics and Mechanics, vol. 12, no. 1-2, pp. 71 – 77, 2000.  C. Clos, “A study of non-blocking switching networks,” Bell System technical journal, vol. 32, pp. 406–424, March 1953.  D. K. Hunter, “Switching systems,” Encyclopedia of Information Technology, vol. 42, pp. 335–370, 2000.  J. S. Turner and R. Melen, “Multirate Clos networks,” IEEE Communications Magazine, vol. 41, no. 10, pp. 38–44, October 2003.  [Online]. Available: http://en.wikipedia.org/wiki/Crossbar_switch  [Online]. Available: http://en.wikipedia.org/wiki/Nonblocking_minimal_spanning_switch  T. Kanai, S. Umemura, S. Inagaki, and S. Hosokawa, “High density pin board matrix switches for automated MDF systems,” IEEE Transactions on Components, Packaging, and Manufacturing Technology, vol. 15, no. 5, pp. 893–903, October 1992.  S. Inagaki, T. Kanai, and S. Hosokawa, “High-density pin-board matrix switch and its crosstalk characteristics,” Electronics and Communications in Japan, vol. 78, no. 7, pp. 42–52, July 1995.  H. Isobe, M. Ito, and T. Murakami, “Automatic MDF (main distribution frame) apparatus,” US Patent 6295483, September 2001.  H. Hosaka, H. Kuwano, and K. Yanagisawa, “Electromagnetic microrelays: Concepts and fundamental characteristics,” Sensors and Actuators, A: Physical, vol. 40, no. 1, pp. 41 – 47, 1994.  H. Hosaka, H. Kuwano, and K. Murata, “Development of miniature matrix relays for telecommunication systems,” Advances in Information Storage Systems, vol. 10, pp. 275–285, 1998.  G. M. Rebeiz and R. Muldavin, “RF MEMS switches and switch circuits,” IEEE Microwave Magazine, vol. 2, no. 4, pp. 59–71, December 2001.  H. A. C. Tilmans, “MEMS components for wireless communications (invited paper),” in Proceedings EUROSENSORS XVI, September 2002, pp. 1–34.  G. Rebeiz, “RF MEMS switches: status of the technology,” in TRANSDUCERS, Solid-State Sensors, Actuators and Microsystems, 12th International Conference on, vol. 2, 2003, pp. 1726–1729.  P. Farinelli, F. Giacomozzi, G. Mannocchi, R. Marcelli, B. Margesin, P. Mezzanotte, S. Di Nardo, P. Russer, R. Sorrentino, F. Vitulli, and L. Vietzorreck, “RF-MEMS SPDT switch on silicon substrate for space applications,” in Topical Meeting on Silicon Monolithic Integrated Circuits in RF Systems, Digest of Papers., 2004, pp. 151–154. REFERENCES 77  B. Yassini, S. Choi, A. Zybura, M. Yu, R. Mihailovich, and J. DeNatale, “A novel MEMS LTCC switch matrix,” in Microwave Symposium Digest,IEEE MTT-S International, vol. 2, 2004, pp. 721–724.  E. McErlean, J.-S. Hong, S. Tan, L. Wang, Z. Cui, R. Greed, and D. Voyce, “2x2 RF MEMS switch matrix,” IEEE Proceedings on Microwaves, Antennas and Propagation, vol. 152, no. 6, pp. 449–454, December 2005.  S. DiNardo, P. Farinelli, G. F., G. Mannochi, R. Marcelli, B. Margesin, P. Mezzanotte, V. Mulloni, P. Russer, R. Sorrentino, F. Vitulli, and L. Vietzorreck, “RF MEMS based switch matrices for complex switching networks,” in Proc. of Microwave Technology and Techniques Workshop, ESA ESTEC, May 2005.  R. Mansour, “RF MEMS for space applications,” in Proceedings of International Conference on MEMS, NANO and Smart Systems, 2005, pp. 191–192.  M. Daneshmand and R. Mansour, “Monolithic RF MEMS switch matrix integration,” in Microwave Symposium Digest, IEEE MTT-S International, 2006, pp. 140–143.  S. Patton and J. Zabinski, “Fundamental studies of Au contacts in MEMS RF switches,” Tribology Letters, vol. 18, no. 2, pp. 215 – 230, February 2005.  H. Kwon, D.-J. Choi, J.-H. Park, H.-C. Lee, Y.-H. Park, Y.-D. Kim, H.-J. Nam, Y.-C. Joo, and J.-U. Bu, “Contact materials and reliability for high power RFMEMS switches,” in Proceedings of the IEEE International Conference on Micro Electro Mechanical Systems (MEMS), 2007, pp. 231 – 234.  L. Chen, H. Lee, Z. Guo, N. McGruer, K. Gilbert, S. Mall, K. Leedy, and G. Adams, “Contact resistance study of noble metals and alloy ﬁlms using a scanning probe microscope test station,” Journal of Applied Physics, vol. 102, no. 7, pp. 074 910 – 1, October 2007.  H. Kwon, S.-S. Jang, Y.-H. Park, T.-S. Kim, Y.-D. Kim, H.-J. Nam, and Y.C. Joo, “Investigation of the electrical contact behaviors in Au-to-Au thin-ﬁlm contacts for RF MEMS switches,” Journal of Micromechanics and Microengineering, vol. 18, no. 10, p. 105010 (9p), October 2008.  V. Jaecklin, C. Linder, N. de Rooij, J.-M. Moret, and R. Vuilleumier, “Lineaddressable torsional micromirrors for light modulator arrays,” Sensors and Actuators A (Physical), vol. A41, no. 1-3, pp. 324 – 9, 1994.  D. A. Winick, B. E. Duewer, S. Palchaudhury, and P. D. Franzon, “Performance evaluation of micromechanical binary phase-only holographic optical elements,” in 47th Electronic Components and Technology Conference, May 1997, pp. 419– 424.  D. A. Winick, B. E. Duewer, S. Chaudhury, J. M. Wilson, J. Tucker, U. Eksi, and P. D. Franzon, “MEMS-based diﬀractive optical-beam-steering technology,” in Proc. SPIE - Int. Soc. Opt. Eng. (USA), vol. 3276, no. 1, 1998, pp. 81–87. 78 Wafer-level heterogeneous integration of MEMS actuators  B. Cunningham, J. Bernstein, D. Seltzer, and D. Hom, “Micromirror pixel addressing using electromechanical bistability,” in Technical Digest. Solid-State Sensor and Actuator Workshop, 1998, pp. 285 – 7.  G. Zhou, L. VJ, F. S. Chau, and F. E. H. Tay, “Line-addressable digitaldeﬂection programmable micromirror array,” Opt. Lett., vol. 29, no. 22, pp. 2581–2583, 2004.  B. E. Duewer, J. M. Wilson, D. A. Winick, and P. D. Franzon, “Programmable MEMS capacitor arrays,” in Proceedings of the SPIE, vol. 3893, no. 1, 1999, pp. 262–271.  A. Jourdain, P. De Moor, K. Baert, I. De Wolf, and H. Tilmans, “Mechanical and electrical characterization of BCB as a bond and seal material for cavities housing (RF-)MEMS devices,” Journal of Micromechanics and Microengineering, vol. 15, no. 7, pp. 89 – 96, July 2005.  J. Thurn and R. Cook, “Stress hysteresis during thermal cycling of plasmaenhanced chemical vapor deposited silicon oxide ﬁlms,” Journal of Applied Physics, vol. 91, no. 4, pp. 1988 – 92, February 2002.  J. Thurn, R. Cook, M. Kamarajugadda, S. Bozeman, and L. Stearns, “Stress hysteresis and mechanical properties of plasma-enhanced chemical vapor deposited dielectric ﬁlms,” Journal of Applied Physics, vol. 95, no. 3, pp. 967 – 76, February 2004.  Z. Cao and X. Zhang, “Mechanism of temperature-induced plastic deformation of amorphous dielectric ﬁlms for MEMS applications,” in Proceedings of the 18th IEEE International Conference on Micro Electro Mechanical Systems (MEMS), 2005, pp. 471–474.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project